Developing a Refined Dynamic Roll-Stack Model for Hot Rolling

Strip Crown Prediction: Developing a Rened Dynamic
Roll-Stack Model for the Hot Rolling Process

By
Derek E. Slaughter
Thesis submitted to the Faculty of the
Virginia Polytechnic Institute and State University
in partial fulllment of the requirements for the degree of
Master of Science
in
Mechanical Engineering
Robert L. West
William T. Baumann
Jim Atkinson
May 26, 2009
Blacksburg, Virginia
Keywords: hot rolling, elastic-plastic, nite element, response surface, sensitivity
c Derek E. Slaughter
STRIP CROWN PREDICTION: DEVELOPING A REFINED DYNAMIC
ROLL-STACK MODEL FOR THE HOT ROLLING PROCESS
Derek E. Slaughter
(Abstract)
The steel industry has been producing at plates through the process of hot rolling since the late
1600s. Hot rolling uses a series of rolls to progressively thin a strip of steel to a desired thickness.
In deforming the strip, the rolling process causes variations in thickness across the width of the
strip. These variations are commonly referred to as crown, which is specically the difference in
thickness between the center and edge of a strip. For most applications steel mill clients require
at products, or products with little variation in thickness. Therefore, variations represent wasted
material which must be removed before the plate or sheet can be used in consumer products.
Controlling the atness of the metal strip is a high priority for the hot rolling business.
The purpose of this work was to develop a 3-D dynamic model of the rolling process to sim-
ulate the behaviour of a strip while being rolled and predict its prole. To accomplish this task,
much of the rolling process needed to be modeled. The prole of the strip is a product of the de-
formation of the rolls and frame within a mill stand. Therefore, not only did the geometry of these
components need to be modeled, but the material properties and dynamic motion were required
as well. The dynamic nature of the process necessitated the modeling of the rotation of the rolls
and translation of the strip, aspects of rolling which are not typically simulated.
Five models were developed during the project. The purpose of the rst two models was to
nd the stiffnesses of the roll-stack and stand frame. The roll-stack refers to the rolls and their ar-
rangement. The reference mill fromwhich data was provided used a four-high roll-stack with two
rolls above the strip and two below. The frame that holds the roll-stack, while massive, stretches
when the strip is deformed between the rolls. This stretch changes the position of the rolls af-
fecting the load and deformation of the strip. A lumped-mass model was created to simulate the
dynamics of the roll-stack and frame. When the strip enters the gap between the rolls, there is a
large impact force which causes the rolls to vibrate. The lumped-mass model was used to deter-
mine parameters to bring the system to steady state. The nal two models simulated the entire
rolling process with rotating rolls and moving strip. The 3-D dynamic rolling model was capable
of predicting the strip prole due after exiting the rolls. Two calibrations were used to reduce
model error before running a validation.
The rolling causes thickness variation across the width of the metal strips; therefore, strips are
intentionally rolled thick to meet a minimum thickness. In modern steel mills, specialized control
systems are used to adjust parameters as the steel strip passes through each stand of rolls. Varying
the parameters allows the thickness and prole of the strip to be controlled. Each stand may have
several rolls in different congurations. These rolls are either work rolls, which directly contact
the strip, or backup rolls, which contact the work rolls and stiffen the roll-stack. The stand frame
holds the rolls and provides a means to position them.
The validation results showed that the exit thickness, strip crown, and rolling load were less
than 5%different fromthe values measured in the test data. The calibrated model was then used to
derive strip crown sensitivities to gap, entry crown, work roll crown, and bending force. The 3-D
dynamic model was able to predict the strip crown accurately when given calibrated information
about the system. This model will be a useful tool for exploring the mechanics of hot rolling in
ways that were not previously possible.
Dedication
This work is dedicated to those whom will learn and benet from it.
iii
Acknowledgements
This work would not have been possible without the assistance and dedication of many people.
My advisor, Dr. Robert L. West Jr., was always available to provide encouragement, guidance,
and expertise far beyond the scope of a faculty advisor. Our shared interests in modeling and
simulation provided a foundation for discourse and a creative environment in which to develop
the project. Jim Atkinson and Thomas Ranger of Industrial Process Support Services sponsored
the project and made available equipment to develop and run simulations. Jim and Thomas were
involved throughout the project, and provided me with the opportunity to learn a great deal about
many subjects in which I am interested. My wife, Jenny Slaughter, provided encouragement and
expert editing skills. Finally, I would like to thank my father, Michael Slaughter, who introduced
me to the principles of engineering through the many projects we worked on together.
iv
Contents
Dedication iii
Acknowledgements iv
Contents v
List of Figures ix
List of Tables xi
List of Symbols xii
Uppercase Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii
Lowercase Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Greek Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2.1 Flat Rolled Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.2 Types of Rolling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.3 Rolling Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Strip Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Roughing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Finishing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Measuring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Coiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.4 Mill Stand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Roll Detail . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Roll Conguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Stand Actuators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Frame Stretch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.5 Strip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.6 Flatness Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.7 Finite Element Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Test Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Thesis Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 Scope of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 Organization of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Modeling of the Hot Rolling Process 13
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Strip Material Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
v
Contents
2.2.1 Strip Spring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2 Rigid-Plastic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.3 Elastic-Plastic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.4 Visco-Plastic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Rolling Load Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3.1 1-D Mathematical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Ford and Alexander Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3.2 2-D Dynamic Finite Element Models . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Strip Prole Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.1 Beam Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Simple Beam Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Slit Beam Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4.2 3-D Finite Element Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4.3 3-D Static Finite Element Models . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4.4 3-D Dynamic Finite Element Models . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 Strip Crown Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3 Rolling Model Development 22
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2.1 Abaqus Finite Element Analysis Software . . . . . . . . . . . . . . . . . . . . . 23
3.2.2 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.3 Problem Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.4 Meshing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2.5 Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Spring & Damper Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Continuum Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Numerical Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.6 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.7 Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.8 Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.9 Implicit and Explicit Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.10 Finite Element Analysis Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 Strip Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3.1 Material Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3.2 Strip Stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.4 Model Simplications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4.1 Stand Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4.2 Frame Stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4.3 Strip Displacement Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.5 Static Roll-Stack Stiffness Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5.2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5.3 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.5.4 Analysis & Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.5.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
vi
Contents
3.6 Static Frame Stiffness Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.6.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.6.2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.6.3 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.6.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.7 Dynamic Lumped-Mass Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.7.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.7.2 Single Degree-of-Freedom System . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.7.3 Multiple Degree-of-Freedom System . . . . . . . . . . . . . . . . . . . . . . . . 48
3.7.4 Explicit Method Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7.5 Model Run Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7.6 Lumped-Mass Model Implementation . . . . . . . . . . . . . . . . . . . . . . . 51
3.7.7 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.8 Dynamic 2-D Rolling Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.8.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.8.2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.8.3 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.8.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.9 Dynamic 3-D Rolling Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.9.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.9.2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.9.3 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.9.5 Mesh Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.9.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4 Calibration, Validation, & Sensitivity 73
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.2 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.2.1 Selection of Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.2.2 Calibration Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.2.3 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.3 Yield-Gap Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.3.1 Response Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.3.2 Error Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.3.3 Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.3.4 Simulation Generation & Running . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.3.5 Data Extraction & Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.3.6 Calibration Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.4 Work Roll Crown Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.4.1 Response Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4.2 Data Extraction & Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4.3 Calibration Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.5 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.5.1 Yield-Gap Calibration Validation . . . . . . . . . . . . . . . . . . . . . . . . . . 84
vii
Contents
4.5.2 Final Calibration Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.6 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.6.1 Parameter Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.6.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5 Summary, Conclusions, and Recommendations 90
5.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.1.1 Model Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.1.2 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.1.3 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.1.4 Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.2.1 Predictive Ability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.2.2 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.2.3 Performance versus Effort . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.3 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.3.1 Model Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.3.2 Process Exploration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.3.3 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.3.4 Crown Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Bibliography 96
Appendix: Test Data 98
Appendix: Rolling Load Model 101
viii
List of Figures
1.1 Mill component layout. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Roughing stands. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Finishing stands. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Cooling bed. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Stand conguration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Roll geometry detail. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.7 Types of roll crown. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.8 Division of the roll into core and shell. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.9 Free body diagram of the upper half of a stand. . . . . . . . . . . . . . . . . . . . . . . . . 8
1.10 Strip geometry detail. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.11 Types of strip crown. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.12 Strip centerline thickness change in stand. . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.13 Strip atness error: edge wave and center buckle. . . . . . . . . . . . . . . . . . . . . . . 10
2.1 Rigid-plastic stress-strain curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Elastic-plastic stress-strain curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Simple beam model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Slit beam model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5 Static 3-D nite element roll-stack model . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1 Roll partitioning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Roll meshing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3 Spring and damper elements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.4 Element types. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.5 Example boundary conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.6 Model boundary conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.7 Rolling Model initial conditiions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.8 Elastic-plastic stress-strain curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.9 Strip stress-strain curve used in the FE rolling models. . . . . . . . . . . . . . . . . . . . . 33
3.10 Stand symmetry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.11 Stand stiffness representation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.12 Strip displacement. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.13 Strip displacement curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.14 Eighth model of the roll stack to determine stack stiffness . . . . . . . . . . . . . . . . . . 38
3.15 2-D roll-stack model partition and meshing scheme. . . . . . . . . . . . . . . . . . . . . . 39
3.16 3-D roll-stack model partition and meshing scheme. . . . . . . . . . . . . . . . . . . . . . 40
3.17 Roll-Stack Stiffness Model boundary conditions. . . . . . . . . . . . . . . . . . . . . . . . 40
3.18 2-D roll-stack load-displacement curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.19 3-D roll-stack load-displacement curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.20 Eighth model of the roll stack to determine frame stiffness. . . . . . . . . . . . . . . . . . 42
3.21 Roll-Stack Stiffness Model boundary conditions. . . . . . . . . . . . . . . . . . . . . . . . 43
ix
List of Figures
3.22 2-D frame stiffness load-displacement curve. . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.23 3-D frame stiffness load-displacement curve. . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.24 Lumped mass model of a quarter roll stack. . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.25 Mass-spring-damper system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.26 single degree of freedom system response. . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.27 Strip exit thickness H
X
versus time t. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.28 Rolling load P versus time t. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.29 Lumped-mass explicit model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.30 Lumped-mass model strip exit thickness H
X
versus time t. . . . . . . . . . . . . . . . . . 54
3.31 Rolling load P versus time t. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.32 2-D rolling model detail. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.33 2-D rolling model mesh. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.34 2-D Rolling Model boundary conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.35 2-D Rolling Model initial conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.36 2-D simulation rolling load P versus time t. . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.37 Strip prole for determination of exit thickness. . . . . . . . . . . . . . . . . . . . . . . . 60
3.38 2-D exit thickness versus strip length. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.39 3-D rolling model mesh. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.40 3-D Rolling Model boundary conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.41 3-D Rolling Model initial conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.42 3-D rolling model strip detail. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.43 3-D rolling model load versus time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.44 3-D rolling model exit thickness versus strip length. . . . . . . . . . . . . . . . . . . . . . 67
3.45 Sample representation of strip exit thickness versus width. . . . . . . . . . . . . . . . . . 68
3.46 3-D rolling model strip exit thickness versus width. . . . . . . . . . . . . . . . . . . . . . 68
3.47 3-D rolling model exit thickness convergence data. . . . . . . . . . . . . . . . . . . . . . . 69
3.48 3-D rolling model crown prole convergence data. . . . . . . . . . . . . . . . . . . . . . . 69
3.49 Mesh convergence, exit thickness versus number of elements. . . . . . . . . . . . . . . . 70
3.50 Mesh convergence, t quality versus number of elements. . . . . . . . . . . . . . . . . . 71
4.1 Isoparametric transformation of error space. . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2 Rolling load versus time for yield-gap calibration simulations . . . . . . . . . . . . . . . 78
4.3 Exit thickness versus strip length for yield-gap calibration simulations. . . . . . . . . . . 79
4.4 Exit thickness versus strip width for yield-gap calibration simulations. . . . . . . . . . . 80
4.5 Rolling-load error contour,
2
p
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.6 Exit-thickness error contour,
2
h
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.7 Total error contour,
t
=
2
p
+
2
h
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.8 Exit thickness versus strip width for work roll crown calibration. . . . . . . . . . . . . . 83
4.9 Strip crown error versus work roll crown. . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.10 Strip exit crown versus gap. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.11 Strip exit crown versus strip entry crown. . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.12 Strip exit crown versus work roll crown. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.13 Strip exit crown versus bending force. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
x
List of Tables
3.1 Roll-Stack Stiffness Model setup data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Roll-stack model stiffness results and predictions. . . . . . . . . . . . . . . . . . . . . . . 40
3.3 Static Frame Stiffness Model setup data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4 Frame stiffness model results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.5 Lumped-Mass Model setup data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.6 Lumped-mass model results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.7 2-D Rolling Model run data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.8 2-D Rolling Model setup data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.9 2-D Rolling Model results summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.10 3-D Rolling Model run data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.11 3-D Rolling Model setup data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.12 Mesh convergence data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.13 3-D Rolling Model results summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.1 Yield-Gap Calibration run data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2 Yield-Gap Calibration work roll data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3 Yield-Gap Calibration backup roll data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.4 Yield-Gap Calibration strip data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.5 Yield-Gap Calibration stand data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.6 Yield-Gap Calibration design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.7 Yield-gap calibration simulation data summary. . . . . . . . . . . . . . . . . . . . . . . . 79
4.8 Roll crown calibration simulation data summary. . . . . . . . . . . . . . . . . . . . . . . . 83
4.9 Yield-gap calibration validation data summary. . . . . . . . . . . . . . . . . . . . . . . . . 84
4.10 Final validation data summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.11 Sensitivity analysis parameter perturbation summary. . . . . . . . . . . . . . . . . . . . . 86
4.12 Sensitivity analysis results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.1 Yield-gap calibration validation data summary. . . . . . . . . . . . . . . . . . . . . . . . . 92
5.2 Final validation data summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.3 Sensitivity analysis results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
1 Work roll data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
2 Backup roll data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3 Strip data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4 Stand data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
xi
List of Symbols
Uppercase Symbols
C
cr
Critical damping matrix
C Damping matrix
C
Change in strip crown within a stand [mm]

C
E
Strip crown entering a stand [mm]
C
s
Strip crown [mm]
C
w
Work roll diameter crown [mm]
C
X
Strip crown exiting a stand [mm]
D
c
Roll barrel centerline diameter [mm]
D
e
Roll barrel edge diameter [mm]
D
n
Roll neck diameter [mm]
E Elastic modulus [tonne/mm
2
]
F Stand reaction force [tonne]
F Force [tonne]
F
n
Nodal force at time step n [tonne]
H
Change in strip centerline thickness within a stand [mm]

H
c
Strip center thickness [mm]
H
e
Strip edge thickness [mm]
H
E
Strip centerline thickness entering a stand [mm]
H
f
Strip thickness at feather distance in from edge [mm]
H
g
Roll gap [mm]
H
X
Strip centerline thickness exiting a stand [mm]
J Bending force [tonne]
J
b
Balance bending force [tonne]
K Stiffness matrix [tonne/mm]
M Lumped mass matrix
P Rolling load or roll separation force [tonne]
R Work roll radius undeformed [mm]
R
Work roll radius deformed [mm]

R
2
Coefcient of determination
0
The tonne unit refers to metric tonnes force
xii
List of Symbols
S
y
Yield stress [tonne/mm
2
]
S
y
1
Initial yield stress [tonne/mm
2
]
S
y
2
Ultimate yield stress [tonne/mm
2
]
T Period of oscillation [s]
Lowercase Symbols
a
n
Nodal acceleration at time step n [mm/s
2
]
c Damping coefcient
c
f
Frame damping coefcient
c
r
Roll-stack damping coefcient
d Displacement at time t [mm]
d
t
Total displacement [mm]
k Mean ow stress in pure shear
_
S
y
/
3
_
k Spring stiffness [tonne/mm]
k
f
Frame stiffness [tonne/mm]
k
r
Roll-stack stiffness [tonne/mm]
k
s
Strip spring stiffness [tonne/mm]
k
t
Total stiffness of frame and roll-stack system [tonne/mm]
k
x
Strip exit modulus [tonne/mm]
L
p
Arc of contact length between work roll and strip [mm]
l
b
Roll barrel length [mm]
l
n
Roll neck length [mm]
l
s
Strip length [mm]
m Mass
m
b
Backup roll mass [tonne-s
2
/mm]
m
w
Work roll mass [tonne-s
2
/mm]
t Time [s]
t
d
Displacment time [s]
t
r
Run time [s]
u Nodal displacement [mm]
u Nodal velocity [mm/s]
u
n
Nodal displacement at time step n [mm]
v
n
Nodal velocity at time step n [mm/s]
w Strip width [mm]
w
f
Feather width [mm]
xiii
List of Symbols
Greek Symbols
Mass matrix proportional damping coefcient
d
Displacement time coefcient
r
Run time coeffcient
t
Total displacement time [s]
Stiffness matrix proportional damping coefcient
t Explicit time step [s]
Engineering Strain [mm/mm]
1
Yield strain [mm/mm]
2
Strain at end of stress-strain curve [mm/mm]
t
True strain [mm/mm]
Error [%]
c
Crown error [%]
h
Exit thickness error [%]
p
Rolling load error [%]
t
Total error [%]
Isoparametric coordinate
yield stress coefcient
Poissons ratio
i
Modal frequency i = 1, 2 [rad/s]
n
Natural frequency [rad/s]
i
Isoparametric interpolation functions i = 1, 2, 3
Engineering Stress [tonne/mm
2
]
t
True Stress [tonne/mm
2
]
Isoparametric coordinate
Damping ratio
i
Modal damping ratio i = 1, 2
xiv
1 Introduction
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Test Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Thesis Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 Scope of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 Organization of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.1 Motivation
Though rolling steel into at sheets is an established manufacturing process, the steel industry is
perpetually looking for ways to improve the quality of its products. The simplest way to improve
quality is through better understanding and control of the rolling processes. For this reason, mod-
els of the rolling process have been developed throughout history to predict how aspects of the
nal product, such as thickness and atness, may be affected based on the conguration of the
mill. Recent advances in nite element analysis and computing have made it possible to simulate
the complex rolling process in great detail.
All at rolled products experience some atness error due to the manufacturing process. This
atness error can range from small thickness variations to waves in the product itself. These
variations force customers to request thicker material than necessary to ensure that they receive
product of a certain thickness. As many customers need a product of a precise thickness, the
excess must be machined away, wasting valuable time and energy. By reducing variability in
thickness and better controlling the production process, it will be possible to reduce costs and
increase manufacturing efciency.
The development of highly accurate models of the rolling process will lead to the more efcient
production and use of steel. If one is able to predict exactly howchanging a mill conguration will
affect the atness of the product, then a mill could be optimized to minimize atness error and
perhaps eliminate it. This optimization impacts not only the mills, which could produce higher
quality steel, but also the customers who would not need to machine at products before use. In
turn, the optimization would impact all consumers of products containing at rolled steel.
1.2 Background
Flat rolling is the process by which a piece of metal, commonly called a strip or slab, is reduced in
thickness to create at sheets or plate. There are many processes and components used to create
the nal shape of a rolled product. To model the rolling process, the parts and their inuence
must be dened and understood. The following presents the relevant processes and components
necessary to develop a model of the rolling process.
1
Chapter 1. Introduction 1.2. Background
1.2.1 Flat Rolled Products
Flat rolled products can be classied into two groups: plates and sheets. The actual designation
depends solely on the thickness to width ratio which is slightly higher for plate than sheet prod-
ucts. However, the width to thickness ratio is much less than one for all at rolled products, and
the rolling process is the same.
To create a at rolled product, a strip of steel with a given initial thickness is successively
reduced in thickness by being pulled through a series of rollers supported by stands. Each pass
through a stand decreases the strips thickness. The amount of thickness reduction at each stand
is closely controlled to determine the nal thickness of the strip and thus the nal dimensions of
the product. The quality of the product is determined by measuring the amount of variation in its
thickness and accuracy in obtaining nominal thickness [14].
1.2.2 Types of Rolling
The process by which the thickness of a strip is reduced can be executed while strip is cold or hot,
called cold rolling and hot rolling respectively. Cold rolling is a type of at rolling in which the
temperature of the metal being rolled is roughly less than half its melting point. This method does
not require the material to be heated and produces a strip with a better nish and dimensional
accuracy than hot rolling. Cold rolled materials can be stronger as strain hardening occurs during
the rolling process [14] [4].
Hot rolling is used to limit strain hardening in the strip. By plastically deforming the material
while above its recrystallization temperature, the deformed grains are able to return to a stress free
state. In addition, the workability of steel improves with temperature increase because ductility
increases and yield stress decreases [31].
1.2.3 Rolling Process
The process of creating a nished at rolled product from the original slab consists of several
stages. These stages in a typical hot rolling mill are strip preparation, roughing, nishing, mea-
suring, and coiling. A sample hot rolling mill layout is shown Figure 1.1 [14].
Roughing
Stands
Finishing
Stands
Furnace
Runout
Table
Coiler X-Ray Gage
Figure 1.1: Mill component layout.
Strip Preparation
In a rolling mill the strip starts out as a slab of steel in one of two ways. In one case, the strip is
cast in a separate process or location, transferred to the mill, and heated to rolling temperature
before entering the roughing stands. A reheating furnace is used to bring the strip up to rolling
temperature, approximately 1300
Cfor most steels. Alternately, the strip can be cast directly as its
2
entering the mill by a process called continuous casting. In this process, the strip goes frommolten
steel to slab dimensions at the beginning of the mill itself and requires no additional heating. The
thickness of the strip is typically between 200 and 300mmwhen entering the roughing stands [14].
Roughing
The roughing process is carried out in several stands to reduce the strip thickness to approxi-
mately 50mm as well as adjust the width before entering the nishing stage. The roughed width
is approximately the width of the nal product [14]. Figure 1.2 shows two roughing stands at a
steel mill in Indiana.
Figure 1.2: Roughing stands (Photograph by Uwe Niggemeier) [19].
Finishing
The nishing process gives the strip its nal dimensions. As the hot strip travels between the
stands its thickness is successively reduced. Due to the strips thickness being relatively smaller
than its width, the strips increase in width is negligeable. Since this is the stage where the nal
product geometry is determined, the model was developed around the test data collected in this
stage. The strip size entering the nishing stands has little bearing on the exit size other than
width. Also, the strip dimensions do not change signicantly after exiting the nishing stands
other than through cooling effects. Figure 1.3 shows several nishing stands in a steel mill.
Measuring
The strip thickness and atness is measured in real time by a X-ray gage at then end of the n-
ishing stands as shown in Figure 1.1. Measuring the nal dimensions of the strip is vital for the
mill controllers. The controllers adjust mill parameters in real time with feedback from the gage
to minimize strip atness.
3
Figure 1.3: Finishing stands (Photograph by Uwe Niggemeier) [19].
Coiling
With the strip rolled to nominal dimensions, it is run out on rollers and allowed to cool. Once the
product cools sufciently, it is coiled for shipping. A mill cooling bed is shown in Figure 1.4; the
coiler is not shown.
Figure 1.4: Cooling bed (Photograph by Uwe Niggemeier) [19].
4
1.2.4 Mill Stand
Mill stands are composed of the set of rolls between which the strip passes and the structure that
supports those rolls. Stands have already been shown in Figures 1.3 and 1.2, but a more detailed
examination is necessary. A stand like those pictured measures several stories tall. Whether it is
used for roughing or nishing, the stand is the primary way to deformthe strip. Since the nishing
stage determines the nal shape of the strip, the nishing stands are of the most importance.
Figure 1.5 shows a single nishing stand and a graphical representation of all its parts. Only a
portion of the stand in the photograph is visible as it extends below the ground.
Strip
Work Roll
Backup Roll
Frame Bending Actuator
Gap Actuator
Figure 1.5: Stand conguration. (Left) Side view of a nishing stand [19]. (Right) Front and side view detail
of stand components.
The stand is broken down into several parts: rolls, frame, and actuators. The rolls can be
categorized as work rolls and backup rolls, which have similar features though they differ in size.
There are generally two actuators in each stand, one to control the gap through which the strip
passes, and the other to control the bending of the rolls. The arrangement of the bending actuators
can be different from those shown in Figure 1.5 as detailed in section 1.2.4.
Roll Detail
Rolls consist of two parts: barrel and neck. The barrel is the part of the roll that comes in contact
with the strip or another roll. The neck is the thinner part of the roll, part of which rests in a
bearing. The bearing is held by a chock which is connected to the stand frame. The geometry of
the roll is described by the measurements in Figure 1.6.
The necks of the rolls may have a variety of shapes as suited to the bearings of a particular
mill. The barrel, on the other hand, is determined during stand setup and can be different for each
roll in each stand. The simplest form of roll barrel curvature is created by forming a symmetric
parabola from the center of the roll barrel to its edge. The diameter crown C
r
of the roll barrel is
dened as the difference in size between the barrel center diameter D
c
and barrel edge diameter
D
e
as given in Equation 1.1.
C
r
= D
c
D
e
(1.1)
5
The roll crown in Figure 1.6 is positive because the center of the roll has a greater diameter than
the edge. The roll crown may also be neutral or negative as shown in Figure 1.7.
Figure 1.6: Roll geometry detail.
Positive Crown
Neutral Crown
Negative Crown
Figure 1.7: Types of roll crown.
The curvature of the roll barrel affects the way the roll bends and interacts with the strip. By
grinding different proles on the roll barrel, it is possible to change many aspects of the rolling
process.
The roll composition and material also affects the rolling process. The stiffer the roll, the less it
deforms. Rolls can be either solid or composite, such that the barrel has a thin layer on the outside
which has different material properties than the core. Figure 1.8 shows the division of the roll
into shell and core. Rolls are commonly made of steel which can be modeled by a linear elastic
isotropic material represented by an elastic modulus E and Poissons ratio .
Roll Conguration
Another stand aspect that affects the rolling process is the roll conguration. The rolls in a stand
can be broken down into two categories: work rolls and backup rolls. Work rolls directly contact
the strip, while backup rolls support and stiffen the work rolls. Work rolls and backup rolls may
6
Shell Core
Figure 1.8: Division of the roll into core and shell.
be used in different arrangements for specic purposes. The four-high roll-stack conguration
uses two work rolls and two backup rolls as shown in Figure 1.5. There also exist twenty-high
roll-stack congurations that make use of two work rolls and eighteen backup rolls. Figure 1.5
details the rolls and frame in a four-high roll-stack. The mill used for model development used
the four-high roll-stack arrangement [4].
Stand Actuators
As previously stated, stands are composed of rolls and a frame. This frame contains actuators
that position and sometimes bend the rolls. The main actuators, referred to as gap actuators in
Figure 1.5, change the distance between the work rolls. The vertical distance between the work
rolls, known as the roll gap or gap, denoted by H
g
, is the space through which the strip passes. By
actuating the gap it is possible to control the thickness of the strip as it exits the stand.
The secondary actuators are called bending actuators. The bending actuators may be arranged
differently than in Figure 1.5 where they act between the work roll necks. The bending actuators
may also apply force between the work roll and backup roll necks or between the backup roll
necks themselves [27]. This work uses the actuator conguration in which the actuators are be-
tween the work rolls. Bending actuators apply a force to the work roll necks, causing themto bend
around the backup rolls. The force the bending actuators apply is commonly called the bending
force and is denoted by J. The bending actuators also provide a balance force to ensure that the
work rolls and backup rolls remain in contact. Most bending actuators can only apply force in
one direction. To simulate a negative bending force, a default load is applied when the stand is
calibrated called the balance bending force represented by J
b
.
Frame Stretch
In a hot rolling mill the rolling loads necessary to deform the strip may be in excess of 2000 metric
tonnes force
1
. Not only is this force sufcient to deform the strip, but it also deforms the stand
frame. Though the frame is very large, as can be seen in Figures 1.5, 1.2, and 1.3, it stretches
slightly when loaded. This stretch allows the rolls to move apart, changing the gap. As the gap
increases, the load on the strip decreases. Thus, a small deformation of the stand structure can
cause a large change in the deformation of the strip. This deformation must be accounted for
when modeling the rolling process.
A distinction in the load causing the stretch must be made. The force that the rolls exert on the
strip is not the same load that the stand must resist due to the addition of the bending force on the
1
All subsequent references to tonnes refer to metric tonnes force.
7
work rolls. Figure 1.9 shows a free body diagram of the upper half of a roll-stack in a stand. The
stand reaction force F, the rolling load from deforming the strip P, and the bending force J are
also shown.
Figure 1.9: Free body diagram of the upper half of a stand.
As shown in Figure 1.9 the rolling load, reaction force, and bending force is related as in Equa-
tion 1.2.
F = P +J (1.2)
The stand stretch is calculated from the stiffness of the frame k
f
and the stand reaction force F as
shown in Equation 1.3.
Stand Stretch =
F
k
f
(1.3)
1.2.5 Strip
The purpose of the previously described components is to control the geometry of the strip exiting
each stand and the mill itself. The shape or prole of the strip can be inuenced at each stand to
achieve the target dimensions of the nal sheet or plate product. The prole of the strip is typically
described by the measurements shown in Figure 1.10.
Figure 1.10: Strip geometry detail.
The width of the strip is denoted by w. The thickness of the strip along its vertical centerline
is H
c
. The thickness of the strip along the edge is H
e
. Due to the change in width of the strip as it
passes through the various stands, an accurate thickness cannot be measured exactly at the edge.
A feather width w
f
is dened from the edge of the strip to where the measurement can be reliably
8
taken. The thickness at the feather is denoted by H
f
. Similar to the rolls, the strip prole is also
measured by crown. Equation 1.4 denes the strip crown C
s
C
s
= H
c
H
f
(1.4)
as the difference in thickness between the center of the strip H
c
and the feather H
f
. As with the
roll crown, it is possible to have positive, neutral, and negative strip crown. Positive crown occurs
when the feather thickness is less than the center thickness. Neutral crown is when the feather and
center thicknesses are equal. Negative crown occurs when the center is thinner than the feather
thickness. Figure 1.11 shows these three cases.
Positive Crown
Neutral Crown
Negative Crown
Figure 1.11: Types of strip crown.
Since the purpose of a stand is to change the shape of a strip, the model must consider the
change in thickness and crown. Thickness change H
, called draft, is the difference between the

strips entry thickness H
E
and exit thickness H
X
as dened in Equation 1.5 and shown in Figure
1.12.
H
= H
E
H
X
(1.5)
Figure 1.12: Strip centerline thickness change in stand.
Similarly, the entry crown is represented by C
E
and the exit crown by C
X
. Using these terms,
Equation 1.6 denes the change in crown C
within a single stand.

C
= C
X
C
E
(1.6)
9
The prole of the strip is typically assumed to be parabolic in shape. This allows the entire prole
of the strip to be described by very few numbers. For instance, given crown, centerline thickness,
and width, it is possible to construct the entire shape of the strip.
Little is known about the material properties of a strip as it moves through the stands. Empir-
ical models have been developed to describe steel properties at elevated temperatures; however,
materials other than steel, such as aluminum or stainless steel, are also worked in these mills. In
some rolling models the material is assumed to be rigid-plastic, while other models make more
detailed assumptions. The strip material will be discussed further in the literature review.
1.2.6 Flatness Error
Ideally, a nal product coming out of a mill would be perfectly at across its width and length.
Crown represents atness error across the width of the strip, but error can also occur along the
length of a strip. Equation 1.6 describes the change in strip crown that occurs between entering
and leaving a stand. The strip can only tolerate so much change in crown before it develops
another kind of atness error. If the change in thickness at a stand is not nearly uniform, then
it causes the strip to lengthen unevenly across its width. This change in length can cause either
edge wave or center buckle as shown in Figure 1.13. Edge wave occurs when the strip crown
increases too much within a stand and center buckle occurs when the strip crown decreases too
much within a stand. This rule says that the strip crown can only be changed so much within a
stand before creating additional atness error besides prole atness error. Flatness error along
the width and length of the strip must be controlled to create a usable product.
Figure 1.13: Strip atness error: edge wave and center buckle.
The only way to control the atness error is to precisely control the strip crown. The strip
crown is affected by the geometry of the rolls, the gap between the rolls, roll stiffness, roll bending,
stand stiffness, and the strip material properties. While rolling the strip it is only possible to
control a few of these parameters. The essence of controlling strip crown lies in determining
which stand parameters have the greatest effect on crown and how they can be controlled. The
purpose of models in hot rolling is to predict strip characteristics prior to rolling the strip based
on information about the mill.
1.2.7 Finite Element Analysis
Finite element analysis (FEA) is a numerical process by which the solution to partial differen-
tial equations can be found over irregularly shaped domains. Developing this process began in
10
Chapter 1. Introduction 1.3. Test Data
the 1950s when computers became widely available. Since that time, great improvements have
been made in FEA and in computers, allowing faster solutions to more complex problems. FEA
has been applied to the hot rolling process before; however, with advances in computing speed,
only recently has it become feasible to create models that simulate so many aspects of the rolling
process. The nite element analysis software Abaqus was used extensively for the creation of
models in this project. Abaqus is capable of solving the nonlinear contact and material problems
presented by the hot rolling process [22, 29].
1.3 Test Data
The sponsor of this project, Industrial Process Support Services (IPSS) provided test data from
a running mill being used to make at steel sheets. This data included stand frame calibration
data and information from online controllers for the stands in the nishing stage of the mill. The
nishing stage was composed of seven stands with loads, gap, and roll geometry data for each.
Measurements of the nal product thickness and crown were taken after the last stand. The data
from one strip was selected and the effect of stand seven on that strip was used for model de-
velopment and verication. The strip from which the test data was generated was rolled directly
after a roll change operation, so the rolls were initially at room temperature. The data from this
strip was used as it was believed that the rolls would exhibit little thermal expansion due to heat
transfer while rolling. The data collected from stand seven is provided in Appendix: Test Data.
1.4 Thesis Objectives
The objective of this work is to create a dynamic nite element model capable of predicting rolling
load and strip prole, specically strip exit thickness and strip crown. This model could then be
used to determine how changing stand parameters affects the crown of the strip. To achieve
these objectives, the model must be able to accurately simulate the rolling process in three di-
mensional space and predict rolling load, strip exit thickness, and strip crown within 10% error.
This level of modeling requires large amounts of numerical computation; however, performing
3-D simulations is the only way to predict strip crown. In addition, the model must encompass
the dynamic effects of the rolling process such as the oscillation of the rolls and determination
of steady-state behaviour. Dynamic simulation is required to exactly simulate the roll-strip inter-
action from which the pressure distribution and resulting strip and roll deformations are found.
The material of the strip must also be modeled to create a representative simulation of the process.
The model will be considered a success if it can predict the exiting crown of the strip to within ten
percent of measured test data. Finally, the model will be used to perform a preliminary sensitivity
analysis to determine the effect of several parameters on strip exit crown.
1.5 Scope of Thesis
The work presented in this thesis will cover the development of the nite element model and the
methods used to calibrate the given model such that it can predict strip crown in accordance with
test data. Model development will encompass not only the construction of the nal 3-D nite
element model, but also all of the models leading up to it as listed below:
Static Roll-Stack Model
Static Frame Stiffness Model
11
Chapter 1. Introduction 1.6. Organization of Thesis
Dynamic Lumped-Mass Model
Dynamic 2-D Rolling Model
In addition to the development of the models, two calibration methods were used to remove
uncertainties in the test data: yield-gap and work roll crown calibration. The results from these
calibrations were used to create a simulation that validates the model. Once the model was vali-
dated, a preliminary sensitivity anaylsis was performed to determine how several stand parame-
ters affect the strip crown.
1.6 Organization of Thesis
This thesis is divided into ve sections. The rst section deals with previous attempts to model
the rolling process and provides background information which is used in this model. The deriva-
tion of rolling force and strip material models is presented along with a representation of the strip
crown. A comparison of online and ofine models is performed with the advantages and disad-
vantages of each.
The second section consists of model development, detailing the methods used and results
from the models listed in section 1.5. First, the stand is broken down into component parts and
the stiffnesses of the roll-stack and frame are determined via the static nite element models. This
information is then used along with the rolling force model to develop the lumped-mass model to
simulate the dynamics of the roll-stack. The results of the lumped-mass model are then veried
by the 2-D rolling model, which is also used to give insight into the nite element modeling of the
rolling process. Finally, the 3-D rolling model is developed which is the goal of this thesis.
The third section takes the fully developed 3-D rolling model and calibrates it against real
world test data. Two calibrations are considered. The yield gap calibration reduces exit thickness
and rolling load error due to uncertain measurements of strip yield stress and roll gap. The work
roll crown calibration corrects for thermal expansion of the roll changing the roll crown which
affects the strip crown. After each calibration a validation is performed to show the calibration
effects. Finally, the last validation is compared to the test data to assess the models performance.
The fourth section uses the calibrated 3-D rolling model to determine the sensitivity of the
strip crown to several stand parameters. The parameters gap, strip entry crown, work roll crown,
and bending force were used for the sensitivity analysis. The parameters were then ranked based
on the amount they inuence the strip crown.
The fth section discusses the results of the model development and calibration. In addition,
areas of further development are proposed along with ways in which the model can be used for
further research on the topic of hot rolling.
12
2 Modeling of the Hot Rolling Process
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Strip Material Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Rolling Load Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Strip Prole Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Strip Crown Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1 Introduction
Though the process of hot rolling began in 16th century, it was not modeled until 1925 [23]. Early
models relied on simple mathematical solutions, while more modern approaches use complex
numerical methods. Each model relies upon a representation of the strip, either with a stiffness
approximation or a full constitutive model. Models of the rolling process may be broken down
into two categories: rolling load models or strip prole models. Force models are used to predict
rolling load and torque, whereas prole models attempt to calculate the deformed shape of the
strip. Analytical solutions and nite element methods can be used in the solution of both types of
models [9]. The purpose behind modeling the rolling process is to control it, so strip control and
parameter sensitivity are also of importance. This chapter will discuss methods of modeling the
rolling process and factors affecting the strip crown.
2.2 Strip Material Models
The success of any rolling process model is dependent on the representation of the strip properties.
These representations include the following models: strip spring, rigid-plastic, elastic-plastic, and
visco-plastic.
2.2.1 Strip Spring
The simplest representation of the strip is achieved by reducing the material properties to a spring
stiffness. This method was developed by Vladimir Ginzburg in his book High-Quality Steel Rolling:
Theory and Practice. Ginzburg [9] used strip stiffness in a 3-D nite element strip prole model.
The strip modulus k
s
is dened in Equation 2.1
k
s
=
P
wH
(2.1)
where P is the rolling load, w is the strip width, and H
is the change in thickness of the strip [9].

While the simplicity of this representation is appealing, it does not represent the strip well. The
yielding of the material is not modeled; therefore the pressure distribution between the work roll
and strip would be very much an approximation.
13
Chapter 2. Modeling of the Hot Rolling Process 2.2. Strip Material Models
2.2.2 Rigid-Plastic
Many force models use the rigid-plastic model, as do some nite element models. The yield stress
represents all of the material characteristics including all effects of strain rate and temperature.
Figure 2.1 shows a stress-strain curve for a rigid-plastic material. In many cases involving large
strains, the rigid-plastic model is adequate for calculating rolling loads [18, 13].
Figure 2.1: Rigid-plastic stress-strain curve.
Rigid-plastic material models of the strip are used because they are simple and efcient for
calculation. However, this model does not well represent the actual material of the strip. In some
cases, neglecting the elastic response of the strip can have signicant effects on the outcome of the
simulation. For this reason, the rigid-plastic material model was not used in this project [13].
2.2.3 Elastic-Plastic
A further development of the rigid-plastic model is the elastic-plastic model. The elastic-plastic
model is dened by an elastic modulus E, an initial yield stress S
y
1
, and a nal yield stress S
y
2
as shown in Figure 2.2. This allows the material to have elastic deformation followed by plastic
deformation with strain hardening. A variation of this material model was used for the strip in
the development of this projects 3-D rolling model [13].
Figure 2.2: Elastic-plastic stress-strain curve.
14
Chapter 2. Modeling of the Hot Rolling Process 2.3. Rolling Load Models
2.2.4 Visco-Plastic
The visco-plastic material model has the greatest ability to represent the actual strip material. The
stresses in the strip are dependent on the strain rate during deformation. By assuming a constant
yield stress, other material models neglect changes in strain rate during the rolling process. This
model can capture these variations [13]. However, the determination of the material properties to
capture these effects were deemed beyond the scope of this project.
2.3 Rolling Load Models
Rolling load models are used to represent the rolling load and torque required to deform a strip
by a given amount [8]. The deformation in these models is considered to be plane strain with
no deformation along the width of the strip. The assumption greatly simplies the model and
allows for fast evaluation of rolling loads by either 1-D mathematical models or 2-D dynamic
nite element (FE) models.
2.3.1 1-D Mathematical Models
The rst mathematical model for the rolling process was developed by von Karman [13] in 1925;
however, it was the later renement and solution of the equations by Orowan [20] in 1943 that
is considered the standard for modeling the process [13]. Orowans model makes use of static
equilibrium equations between the roll and strip as the strip is undergoing deformation. The
distributed pressure along the arc of contact between the roll and strip, as well as the shear forces,
are used to generate the equilibrium equations [13]. Orowan used a graphical method for the
solution of these equations [13]. A simplication of the Orowan model was made by Sims [28]
in 1954 by assuming that the angles in the roll gap are small [13]. A further simplication of
Orowans model was made by Bland and Ford in 1948 by setting the roll pressure equal to the
vertical stress in the strip [13]. In 1964, Ford and Alexander developed a simplied rolling model
that is still used in the industry [4]. Alexander performed a study in 1972 using a FORTRAN
program to numerically integrate and solve Orowans original equations. He then compared the
results to the simplied models made by Sims and Bland and concluded that the analytic models
were unable to calculate the rolling load with a high degree of accuracy [2]. In 1996, Freshwater [6,
7] presented simplied theories of rolling based on further computational study of von Karmans
and Orowans equations.
Limitations
The 1-D rolling load models have many limitations. The models assume a plane strain environ-
ment in which there is no deformation along the width of the strip. This assumption discounts
the ability of the strip to change width or have variations in thickness across the width, making
the models unable to calculate crown. In addition, effects of the rolling load on frame stretch is
neglected. The material model of the strip is assumed to be simple: perfectly plastic with some
elastic effects at the entrance and exit of the rolls [21]. Even with these limitations, the 1-D rolling
models remain a useful tool in quickly estimating rolling loads.
Ford and Alexander Formulation
The Ford and Alexander rolling load model was used in the process of creating the lumped-mass
model by calculating a pseudo strip stiffness. The model depends on the strip properties: yield
15
Chapter 2. Modeling of the Hot Rolling Process 2.3. Rolling Load Models
stress, entry thickness, and exit thickness and work roll properties: radius, elastic modulus, and
Poissons ratio [4].
There exists a relationship between length of the arc of contact between the rolls and strip, the
deformation of the rolls, and rolling load. The rolls are allowed to deform, which changes the arc
of contact, affecting the rolling load. The length of the arc of contact between the rolls and the
strip is dened in Equation 2.2
L
p
=
R(H
E
H
X
)
(H
E
H
X
)
2
4

_
R(H
E
H
X
) (2.2)
where L
p
is the arc of contact, Ris the undeformed radius of the work roll, and H
E
and H
X
are the
entry and exit thicknesses of the strip respectively [4]. The effect of roll deformation is to change
the roll diameter as calculated using Hitchcocks equation
R
= R
_
1 +
16 (1 )
2
EH
P
w
_
(2.3)
where R
is the deformed roll radius, H
is the change in strip thickness, E is the elastic modulus

of the roll, is the Poissons ratio of the roll P is the total rolling load, and w is the width of the
strip [13]. The rolling load P referred to in the previous Equations is developed as
P = kwL
p
_
2
+
L
p
H
E
+ H
X
_
(2.4)
where k is the mean ow stress in pure shear, w is the width of the strip, and L
p
is the length of
contact. The mean ow stress is related to the yield stress of the material by Equation 2.5.
k =
S
y
3
(2.5)
An iterative solution of the force model is required due to the interactions between rolling load,
arc of contact, and deformed roll radius.
2.3.2 2-D Dynamic Finite Element Models
The 2-D dynamic FE rolling load model typically attempts to model the physical motion of the
strip and the resulting deformation. The nite element modeling of the 2-D rolling process is
superior to the 1-D mathematical models because the boundary conditions and materials are rep-
resented more accurately [18]. Several 2-D nite element models of the rolling process have been
created and shown to offer more accurate approximations of rolling load [18, 5].
The model by Mori et al. [18] was developed to look at strip deformation and rolling loads.
The work rolls were assumed to be rigid and the thickness reduction of the strip was specied.
The strip was composed of isoparametric quadrilaterals with four Gauss points for full integra-
tion. Half of the strip was modeled as it moved beneath the rotating work roll. This model also
demonstrated likely deformation of the strip mesh during rolling [18].
The model by Dvorkin et al. [5] in 1997 was a 2-D nite element model developed as a step
towards a 3-D model incorporating roll elasticity, thermal effects, and a strip material model. This
2-Drolling model represents the systemusing plane strain. Arigid-viscoelastic material was used
to dene the strip. Half of the strip and one work roll is modeled with a symmetric boundary
condition on the strip centerline and the roll rotating about its axis. An Eulerian formulation was
16
Chapter 2. Modeling of the Hot Rolling Process 2.4. Strip Prole Models
used such that the mesh remained stationary with the strip owing through it. Once the model
was developed, it was used to determine process sensitivity to several rolling variables [5].
The model developed in this work is similar to the models by Mori et al. and Dvorkin et al. in
certain respects and different in others. Whereas those two models only simulated the rolling of
the work roll, this projects 2-D model represents both the work roll and backup roll. Deformation
of the rolls was not considered in the discussed models, though it was in the one developed for
this thesis. All three models are able to predict rolling load and strip exit height.
2.4 Strip Prole Models
The purpose of a strip prole model is to predict the shape of a strip exiting a stand. These models
attempt to represent deformation across the width of the rolls due to bending and contact. The
interaction between the strip and the rolls is modeled in greater detail than is possible with rolling
load models. Many different representations for the strip are used. Most strip prole models are
based on beam or nite element theory.
2.4.1 Beam Models
Beam models are based on beam theory and can be classied as simple beam models or slit beam
models. The simple beam model represents the entire roll or rolls as a single beam and integrates
the beam equation over the width of the roll using a distributed rolling load to determine deec-
tion. The slit beam model represents the roll as a series of beam segments coupled by inuence
coefcients. By subdividing the rolls, the variation in the loads along the width of the strip can be
better represented through equivalent nodal loads. Beam models are commonly used by online
mill controllers as they are able to quickly predict strip proles while the strip is being rolled [17].
Simple Beam Models
The rst simple beam model was developed by Saxl in 1958. Saxl developed models for two-high
and four-high mills. The two-high model uses beam theory as illustrated in Figure 2.3. The center
of the beam is assumed to be a cantilever while the edge of the barrel is simply supported. The
rolling load is represented as a uniformloading over the width of the strip. To obtain the deection
of the centerline of the roll, the beam equation was integrated over the width of the roll. The
displacement due to bending and shear effects are superimposed to give the total deformation.
Adjustments are also made for contact deformation between the strip and roll. Asimilar approach
is taken with a four-high model using two beams with an elastic foundation. However, the backup
roll is assumed to be rigid, having no centerline deection. Neither of these models is capable of
representing roll crown in their formulation [26, 9].
Figure 2.3: Saxls two-high roll-stack simple beam model [26].
17
A more advanced simple beam model was developed in 2007 by A. Malik and R. Grandhi to
represent twenty-high mills. Beam elements were developed using foundation moduli between
the rolls. The beam elements could represent the barrel or neck of the roll along with variations in
cross section. This model was signicantly more exible than Saxls. The main advantage of this
model was that many rolls could be combined in various congurations to dene the roll stack
[17].
There are several deciencies in the simple beam model. The pressure distribution is not uni-
form in the actual rolling process which leads to different bending behaviour as noted by Shohet
and Townsend [27]. Secondly, the actual rolls behave as though the necks are cantilevered be-
cause the bearings do not allow them to rotate in the bending plane. This boundary condition
leads to signicantly different bending. Finally, 3-D nite element models are able to simulate
the localized deformation in the roll-strip contact region which the simple beam model can only
approximate.
Slit Beam Models
The slit beam model created by Shohet and Townsend [27] in 1968 is a blend of the simple beam
model and nite elements. The rolls and strip are broken down into elements of varying lengths
depending on the roll and strip geometry. Rigid body motion of the rolls is allowed and crown
is modeled. The bending of the rolls is calculated through inuence coefcients and deformation
in the roll-strip contact region is determined on an element by element basis. This representation
allows for a more localized deformation at the edges of the strip. Compatibility conditions resolve
the forces in rolling load and bending [10]. Compatibility conditions are written between the work
roll and backup roll and the work roll and the strip. These conditions allow for the representation
of roll crown by mathematical representing the closing of the gaps that crown creates. The equi-
librium conditions of the rolls and strip allow for the solution of the equations as the sum of the
forces between the rolls and strip must be zero. Figure 2.4 shows the discretization of the backup
roll, work roll, and strip in a four-high slit beam model.
Figure 2.4: Slit beam model of the four-high roll-stack [9].
A more recent slit beam model was developed by Yun et al. which uses a 3-D nite ele-
ment model to determine compliance coefcients between the rolls themselves and the strip. This
method allows for calibration of the model against a static 3-D simulation for additional accuracy
[32].
Slit beam models also have deciencies with respect to 3-D nite element models. The inu-
ence coefcients are based on beam theory which may not be suitable for the rolls, because of the
large diameter to length ratio called the slenderness ratio. The slit beam model also assumes that
18
the work roll and backup roll are in contact across the entire width of the rolls. Finally, the slit
beam model does not completely model the roll-strip contact, which is approximation with dis-
creet spring elements representing an elastic foundation [9]. The 3-D rolling nite element model
better represents the roll geometry and deformation along with the contact and nonlinear strip
properties.
2.4.2 3-D Finite Element Models
Finite element models are the most advanced models capable of predicting strip crown. The
reason nite element models are so accurate is that they are capable of representing the geometry
of the roll and strip in great detail through discretization. Not only is the geometry of the strip and
rolls well represented, but also the material properties can be modeled in detail by choosing any of
the models in section 2.2. Finite element models for predicting strip prole fall into two categories:
static and dynamic. Static nite element only consider applied loads and elastic forces. Dynamic
models include inertial and damping forces in addition to applied loads and elastic forces. There
are advantages and disadvantages to both methods.
2.4.3 3-D Static Finite Element Models
Static 3-D FE models are similar to the simple and slit beam models except they are better able
to represent the geometry and the strip material. A notable static model developed by Ginzburg
was called the ROLL-FLEX
TM
which simulates the behaviour of one eighth of a four-high mill. A
one-eighth roll-stack 3-D static model is shown in Figure 2.5. The eighth model is used due to
symmetry conditions and the assumption that the rolls are not rotating. These simplications,
along with the strip spring model, allowed for a simple nite element model. Because of the
simple strip material model and the lack of rolling, the pressure distribution between the work
roll and strip is an approximation [9].
Figure 2.5: Static 3-D nite element model of a four-high roll-stack [9].
A similar eighth model of a four-high mill was developed by Malik and Grandhi for verica-
tion of their simple beam model [17]. Malik and Grandhi represented the strip using a foundation
19
Chapter 2. Modeling of the Hot Rolling Process 2.5. Strip Crown Control
modulus and rolls discretized with tetrahedral elements [17]. The goal of both the simple beam
and 3-D nite element models was to predict the strip exit crown.
2.4.4 3-D Dynamic Finite Element Models
Dynamic 3-D FE models represent the deformation of the rolls and strip with the greatest accu-
racy. These models simulate the actual rotation of the rolls and translation of the strip. Therefore,
the contact conditions and pressure distributions must develop in the same manner as the actual
rolling process assuming the material and geometry models are accurate. However, these dy-
namic models must also deal with reaching steady state conditions and large runtimes. The nal
model developed for this project is a 3-D dynamic nite element model.
Until recently, fully developed dynamic FE simulations of the 3-D rolling process have been
considered too computationally expensive [11]. This computational requirement has been used as
a justication for the simplications made by other beam models [11]. However, deciencies in
other models lead to the continued development of nite element models [11]. An early dynamic
3-D model of the rolling process was developed in 1997 by Zone-Chin lIn and Ven-Huei Len for
the study of thermal expansion in the work roll [15].
The purpose of the model in this project is to use the 3-D dynamic rolling model to determine
which controllable parameters most affect strip crown and how much that affect is. This work
will perform analysis similar to the work of Shohet and Townsend, discussed in the following
section, using a fully dynamic model to achieve highly accurate results. This level of analysis has
not been previously performed due to prohibitively long computing times. By applying specic
control parameters to the roll stack and using modern computer equipment, the run times have
been reduced to the point where this analysis is feasible.
2.5 Strip Crown Control
The rst studies to control strip crown were performed by Shohet and Townsend in 1968 using
a slit beam model [27]. There are many factors that affect strip crown. Shohet and Townsend
identify several parameters that affect strip crown: roll geometry, draft, rolling load, bending
force, strip temperature, strip entry crown, strip entry thickness, and strip width. Roll geometry
not only includes the sizes of the neck and barrel, but also the roll crown [27].
Certain stand parameters, such as roll size are determined by mill designers or material pro-
cessing requirements and cannot be readily controlled to change strip crown. However, bending
force, rolling load, and strip entry crown can all be inuenced at run time and can have a large
impact on strip crown [27].
The roll crown is determined by the superposition of several factors: ground crown, ther-
mal crown, and wear crown. Shohet and Townsend represented all three types of crown with a
parabola; thus, the combined crowns can be represented by a single parabola [27]. Therefore, by
adjusting the work roll crown it is possible to simulate the effects of thermal expansion and wear.
The thermal crown generated while rolling can be signicant [24]. One experiment showed the
roll crown to increase by 0.30 mm in 52 seconds while rolling a 1270 mm wide strip [9].
The next largest controllable inuence on strip crown is the bending force. The bending force
deforms the work roll around the backup roll which also effectively changes the roll crown. How-
ever, the effects of the bending on the roll crown cannot be represented as another parabolic crown
on the roll.
While it is believed that entry crown has little inuence on the strip exit crown, it still merits
discussion [27]. The entry crown at each stand is the previous stands exit crown. Thus, the entry
20
Chapter 2. Modeling of the Hot Rolling Process 2.6. Summary
crown can be controlled along with its effects at each successive stand.
The sensitivity of the strip crown to gap, entry crown, roll crown, and bending force are be-
lieved to be the primary variables to be controlled by an online controller. A study of crown
sensitivity can show which parameter has the greatest inuence on strip crown and how much
the parameter must be changed to minimize strip crown [27]. Sensitivity can be found by perform-
ing a regression of crown versus varying parameter data. The slope of this regression describes
the rate at which strip crown changes with respect to a given parameter. For instance, Shohet and
Townsend [27] study the effects of bending force on strip crown to nd sensitivity.
2.6 Summary
There exist many components to rolling process modeling. Consideration must be given to model-
ing the strip material, roll-strip contact deformation, and roll bending. Models range from simple
1-D mathematical models to complex 3-D dynamic nite element models. Using these models
to determine the effects of stand parameters on strip crown can be useful in controlling the strip
crown. Parameter sensitivities can be calculated to determine which parameters have the most
effect on strip crown. The nal model of this work will use a 3-D dynamic nite element model to
determine crown parameter sensitivities.
21
3 Rolling Model Development
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Strip Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4 Model Simplications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.5 Static Roll-Stack Stiffness Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.6 Static Frame Stiffness Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.7 Dynamic Lumped-Mass Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.8 Dynamic 2-D Rolling Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.9 Dynamic 3-D Rolling Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.1 Introduction
The nal roll-stack model was the result of several models, each one building upon the last. These
models were developed using the nite element method and a simplied representation of the
stand. Each model was a necessary step forward and a check to verify the results of the previous
model. The models are presented in the following order:
Strip Material Model
Static Roll-Stack Model
Static Frame Stiffness Model
Dynamic Lumped-Mass Model
3.2 Finite Element Method
The Finite Element Method (FEM) or Finite Element Analysis (FEA) was used extensively for the
modeling of the rolling process in this work. FEAwas used to simulate the structural deformation
and dynamics of the roll-stack system. Both 2-D and 3-D FE models were created and used at dif-
ferent stages. Finite element modeling, as in all types of modeling, depends greatly on the choices
of representation for the system being modeled. Each choice has advantages and disadvantages
that can impact the quality of the solution. Many important decisions were made when the nite
element models were created. The aspects of FEA that apply to the models in this project will be
addressed in this section.
22
Chapter 3. Rolling Model Development 3.2. Finite Element Method
3.2.1 Abaqus Finite Element Analysis Software
The FEA software Abaqus, distributed by Simulia, was used for the Roll-Stack, Frame Stiffness,
and Rolling models in this work [29]. Abaqus is designed to solve highly nonlinear material and
contact problems in both static and dynamic models. Abaqus is composed of three products:
CAE, Standard, and Explicit. Abaqus CAE is the graphical interface through which models are
built and updated, as well as where output data is viewed and extracted. In short, CAE handles
the preprocessing and postprocessing of the models. Abaqus Standard and Explicit are types
of solvers; Standard uses the implicit nite element formulation while Explicit uses the explicit
nite element formulation, both methods are discussed in section 3.2.9. Abaqus CAE provides
a programming interface using the Python scripting language. The scripting ability of CAE was
used extensively as a means of updating the models and exploring the effects of different stand
parameters on the solution.
3.2.2 Purpose
The purpose of the nite element method is to solve partial differential and integral equations
over irregularly shaped domains. In the case of structural and dynamic analysis, the type of
analyses performed in this thesis, the systemconsists of a body which can move or deformin space
and time. FEA breaks down the body into regular domains called elements through a process
known as meshing. The characteristics of each element depend on the elements size, shape, and
material properties. This information can be used to calculate the elements stiffness and mass.
The elements are then assembled to represent the entire body of the original system. To dene
how the body will behave, boundary conditions such as displacements and forces are prescribed
on the body. Initial conditions such as velocities or accelerations are also applied. With the system
fully described, it is possible to solve for element displacements and deformations. In addition,
stresses, strains, velocities, accelerations, contact pressure, reaction forces, internal energy, and
kinetic energy can be found. From these results, predictions about how a system behaves can be
gathered and analyzed.
3.2.3 Problem Types
Finite element analysis problems consist of three main types: boundary value problems, initial
value problems, and initial boundary value problems. Boundary value problems describe a spa-
tially dependent, time independent system governed by differential equations whose solutions
are based on boundary conditions. Both the Static Roll-Stack Model and Static Frame Stiffness
Model are examples of boundary value problems because they solve for the time independent
deformation of the rolls. An initial value problem represents a spatially independent, time depen-
dent system which responds over time to a set of equations and initial conditions. The Dynamic
Lumped-Mass Model is an example of an initial value problem because it does not represent the
spatial aspects of the rolls, only the time dependent response. An initial boundary value prob-
lem is a combination of the initial value and boundary value problem resulting in both a spatially
dependent, time dependent representation of a system. The Dynamic 2-D Rolling Model and Dy-
namic 3-DRolling Model are initial boundary value problems because the deformation of the rolls
and strip are found as a function of time.
23
3.2.4 Meshing
Meshing, the rst step in nite element analysis, refers to the decomposition of bodies such as
rolls and strip into subdomains or elements. The quality of a bodys mesh can have a large impact
on the overall quality of the model. The distribution of elements and their sizes affect how well
they are able to represent the mechanics of a body. The purpose of an element is to represent not
only the geometry of a body, but also its stiffness.
The choice of element type and number can inuence how well a bodys geometry is repre-
sented. For instance, a linear element can only represent linear geometry such as at faces or
edges. To model a curved surface such as a roll with linear elements, a large number of elements
must be used. Conversely, a small number of quadratic elements can be used because they can
represent geometry described by second order polynomials. However, quadratic elements have
other limitations which are discussed in Section 3.2.5. The ability of an element to properly repre-
sent geometry is especially important in contact problems where element boundaries in separate
bodies determine the forces acting between the bodies. If too few elements are used, the dis-
cretization of the force between the bodies cannot adequately represent the pressure distribution
in the contact zone.
The shape and type of element also affects how well it is able to represent the stiffness of a
body. Elements with large aspect ratios, the ratio of an elements longest side to its shortest side,
are unable to represent the stiffness of the body as well as a cube or square shaped element. The
inaccurate stiffness stems from the distortion of the stiffness calculated in isoparametric space
and transformed into physical space. A linear element also cannot represent a nonlinear stress
distribution because the stresses within the element vary linearly. This makes large elements
inappropriate in areas where stress does not vary linearly.
If an element or series of element misrepresents the stiffness in a body, the solution will be
invalid. For this reason developing a good mesh is very important. To create a representative
mesh for the rolls, they was divided into two regions: the core and the shell in accordance with
the material composition of the roll. The shell and core sections were further partitioned into areas
in 2-D and volumes in 3-D onto which specic mesh sizes were applied. The partitioning scheme
for the work rolls in both the dynamic rolling and static stiffness models are shown in Figure 3.1.
The lines represent partition and section boundaries.
The shaded regions of Figure 3.1 denote areas of ne and coarse meshing. Fine meshing is lo-
cated where the roll was expected to be in contact with either the strip or another roll. Ane mesh
is necessary in the areas of contact as an accurate representation of the geometry and deformation
is especially important in these areas. For Figure 3.1a, the contact area is large because the roll is
rotating and is in contact with the strip along that surface throughout the simulation. The contact
area in Figure 3.1b is small as the roll is stationary.
Figure 3.2 shows a typical mesh on a work roll. The mesh is carefully constructed so that the
elements in the shell and core are aligned except in the nely meshed areas. A similar technique
was used with the strip though no partitioning was required and smaller elements were used.
The quality of a mesh can be determined by running models with varying meshing sizes and
measuring the effect on a variable, such as stress at a specic location. As the size of the elements
decrease and the number of elements increase, the change in stress should decrease and approach
a constant value. This is known as a mesh convergence test in which a measurement is converged
to within a given percentage. The minimum number of elements that give the required conver-
gence are used in the nal models because additional elements would not signicantly contribute
to the accuracy of the model. This method was employed to determine the quality of the meshes
for this work.
24
(b) (a)
Partition
Boundry
Section
Boundry
Fine Mesh
Coarse Mesh
Figure 3.1: Roll partitioning. (a) Dynamic rolling model. (b) Static roll-stack model.
Core
Shell
(b) (a)
Figure 3.2: Roll meshing. (a) Dynamic rolling model. (b) Static roll-stack model.
3.2.5 Elements
There are many different types of elements that can be used to represent components of a model.
The two groups of elements used in the models for this work were spring and damper elements,
and continuum elements. Spring and damper elements are some of the simplest nite elements
and were used to represent stand stiffness and damping. Continuum elements are more complex
and were used to model the rolls and strip as shown in the meshing scheme in Section 3.2.4.
Spring & Damper Elements
Springs and dampers are special discrete lumped parameter elements for which stiffness and
damping coefcients can be prescribed directly. Spring elements create a coupling between two
nodes such that the relative displacement between the nodes causes a force to be generated op-
posing the direction of displacement. This force is generated per Hookes law, Equation 3.1,
F = k(u
2
u
1
) (3.1)
where F is the internal force in the spring element. If the element is tied to the ground the internal
force will also be a reaction force. The stiffness coefcient is represented by k, u
1
is the displace-
ment of node 1, and u
2
is the displacement of node 2. The spring system is illustrated in Figure
25
3.3. A similar equation can be used for the damper element to relate the damper internal force to
the relative velocities of the two nodes as shown in Equation 3.2.
F = c( u
2
u
1
) (3.2)
The damping coefcient is represented by c, u
1
is the velocity of node 1, and u
2
is the velocity of
node 2. The damper element is shown in Figure 3.3. Spring elements were used in the static and
dynamic models to represent the stand. Dampers were used in the dynamic models to control the
oscillatory response of the roll-stack in the stand and between the rolls themselves.
2 1
2 1
Figure 3.3: Spring and damper elements.
Continuum Elements
While the mesh denes the shape of the elements in a body, the elements have many individual
properties. The type of element and method by which the element stiffness is generated is as im-
portant as the quality of the mesh. If an element of the incorrect formulation is used, it will likely
misrepresent the stiffness of the body. Two types of elements were used in the construction of
the FE models in this work: bilinear quadrilaterals and trilinear hexahedrals or bricks. Bilinear
quadrilaterals are 2-Dplane elements with four nodes and eight displacement degrees of freedom.
The 2-D FE models were comprised of hundreds or thousands of these elements arranged into the
shape of the rolls and strip. Trilinear hexahedrons are the 3-D version of the bilinear rectangle
with eight nodes and twenty-four displacement degrees of freedom. By looking at these two ele-
ments, it is possible to see how the complexity of the solution increases quickly from 2-D to 3-D.
Not only does the number of elements and nodes increase, but so do the number of displacement
degrees of freedom [3]. A diagram of these elements is shown in Figure 3.4.
Both the quadrilateral and hexahedral are linear isoparametric elements, which are only ca-
pable of representing linear variations in geometry and the displacement eld and the result-
ing stresses and strains across the element. This limitation is due to the interpolation functions
dening the element. In the case of the bilinear rectangles and trilinear bricks, the interpolation
functions vary linearly in all of the directions across the element. The isoparametric interpolation
functions in element parametric space or and space for the 2-D element in Figure 3.4 are given
in Equation 3.3 [22].
1
=
1
4
(1 )(1 )
2
=
1
4
(1 + )(1 )
3
=
1
4
(1 )(1 + )
4
=
1
4
(1 + )(1 + )
(3.3)
26
1 5
2
6
7
8
3
4
1 2
4 3
(a) (b)
Figure 3.4: Element types. (a) 2-D bilinear rectangle element. (b) 3-D trilinear brick element.
Elements with quadratic or cubic interpolation functions do exist; however, those elements do
not typically represent distributed contact loads along element faces accurately unless they are
modied for this purpose [3, 1].
The stiffness of an element is calculated by integrating the material stress-strain properties
over the domain of the element to determine the resistance of an element to deformation. Deform-
ing an element by displacing its nodes causes internal stresses and forces which react against the
nodes. For a 2-D quadrilateral element, the stress-strain relationship can be formulated as plane
stress or plane strain. In plane stress, the element thickness is assumed to be very thin relative
to the planar area resulting in out-of-plane stresses of zero. In plane strain, the element thickness
is assumed to be very thick relative to the planar area, which results in out-of-plane strains of
zero. Both the plane-strain and plane-stress stress-strain relationships are approximations of the
actual material state. The 3-Dhexahedral element provides the most accurate representation of the
stresses and strains as they are represented in all directions. For the 2-D FE models, a plane strain
stress-strain relationship was used because the strip width is much larger than the strip height
which results in very small displacements of the strip in the width direction. The 3-D models use
the 3-D hexahedral element, providing the most accurate stiffness representations [16].
Numerical Integration
In FEA, the integration of an elements material properties to calculate its stiffness is typically a
form of numerical integration, or quadrature. The method in which the stiffness is integrated also
impacts the stiffness of the element. The most common formof quadrature for developing element
stiffness is Gauss quadrature. Like all forms of quadrature, Gauss quadrature involves sampling
stiffness at several points throughout the element, multiplying by weights, and summing to nd
the total stiffness. Where these samples are taken, the number of samples, and their weights
are determined by a set of quadrature rules. The number of sample points, or Gauss points,
is determined by the order of the elements interpolation functions. A polynomial of degree p
can be exactly integrated by Gauss quadrature of order n where p 2n 1. Thus, 1-D linear
element would only require one Gauss point to be integrated exactly because the interpolation
functions are rst degree. However, a 2-D linear quadrilateral, described by the interpolation
functions in Equation 3.3, requires four Gauss points, two in each dimension, to integrate the
second degree interpolation functions. Thus, element stiffness can be found exactly and effectively
through numerical integration [3].
27
In some cases it is useful to underintegrate an elements stiffness by using fewer Gauss points.
This method, called reduced integration, occurs when fewer Gauss points are used than the num-
ber required to exactly integrate the functions. For example, the 2-Dlinear element which requires
four Gauss points to be integrated exactly, could be integrated with one Gauss point. Reduced in-
tegration has advantages and disadvantages depending on how it is used.
The main advantage to reduced integration is a large decrease in the time required to compute
an elements stiffness. Because the element stiffness must be evaluated at each Gauss point, going
from four points to one reduces the computation time by 75%. Reduced integration has an effect
on the stiffness of the element. Since the element stiffness is sampled in fewer places, the element
does not have stiffness related to certain modes of deformation. In some cases this lack of stiffness
can cause problems; however, in bending quadrilateral elements tend to be too stiff or to lock
when compared to experimental results. Reduced integration decreases the element stiffness in
bending an so yields more accurate results. For reduced integration of linear elements Abaqus
does not simply sample the element stiffness at a single point. Instead, the stiffness is computed
from the average strain over the volume of the element based on a uniform strain formulation.
The formulation Abaqus uses ensures element stability [3, 1].
Reduced integration elements were used in all of the applicable models developed for this
thesis. The reduction in run time due to using reduced integration is signicant as the element
stiffnesses must be calculated at every step in the nonlinear analysis. Also, reduced integration
elements allow the rolls to bend more accurately.
3.2.6 Boundary Conditions
A boundary condition (BC) is an external condition applied to a system such as a force, pressure,
or displacement. Boundary conditions help to dene how a system will behave when simulated.
The two main types of boundary conditions used in the models for this work are displacement
and load . Displacement boundary conditions describe howelement nodes move and load bound-
ary conditions describe forces on nodes. A displacement and load boundary condition can not be
applied on the same node degree of freedom at the same time, only one or the other may be pre-
scribed. Figure 3.5 uses beam elements to illustrate displacement and load boundary conditions.
a
b
1 2 3
1
2
3
1 2
1
2
Figure 3.5: (a) Fixed-xed beam undeformed and deformed. (b) Simplied beam undeformed and de-
formed.
28
Displacement boundary conditions can be prescribed on any degree of freedom for any node
in a mesh. The displacement can tell how far the node will move along a given degree of freedom,
or specify that the node will remain stationary. In Figure 3.5 each node has three degrees of
freedom, horizontal, vertical, and rotational. Nodes 1 and 3 are constrained in all three degrees
of freedom, meaning that they cannot rotate or translate, because this beam is xed on both ends.
When a node is xed in all degrees of freedom is also called an encastre boundary condition.
In Figure 3.5a node 2 has no prescribed displacements and is only affected by the force which
is applied as a load boundary condition to the vertical degree of freedom. However, Figure 3.5a
can be simplied using a symmetry boundary condition. The symmetry boundary condition is
a special form of the displacement boundary condition which constrains the motion of node 2
in Figure 3.5b such that it responds in the same manner as Figure 3.5a. This is accomplished by
prescribing that node 2 cannot displace horizontally, nor rotate. This results in node 2 having the
same displacement in both deformed beam representations shown in Figure 3.5.
Another special displacement boundary condition is used in dynamic models, the velocity
boundary condition. The velocity boundary condition prescribes the rotational or translational
velocity of body, element, or node. A displaced symmetry boundary condition, a combination of
the symmetry and prescribed displacement boundary conditions, is used in several models. The
displaced symmetry boundary condition effectively moves the plane of symmetry along a path
normal to the plane.
The previously mentioned boundary conditions are used throughout the models developed
in this work. For clarity a boundary condition gure is included for each model to describe the
type and location of boundary conditions used in that model. A summary of model boundary
conditions is presented in Figure 3.6. The symmetry boundary conditions indicate that the surface
may not move perpendicular to the plane of symmetry, but may slide freely in that plane. The
encastre boundary condition is used to connect the roll or frame to ground, which does not move.
The horizontal xed boundary allows the roll to translate vertically and rotate, but not move
horizontally which simulates being held in the stand frame. The velocity boundary condition is
applied to the work roll in the Strip Rolling Model because the work roll is driven by a motor in
the actual stand. Lastly the bending force is denoted by a load boundary condition on the end of
the work roll neck.
3.2.7 Initial Conditions
Initial conditions (IC) are boundary conditions that describe the state of the model at the begin-
ning of the simulation. Initial conditions allow the simulation to start with components at rest
or already in motion. In a static analysis dynamic effects are ignored so initial conditions do not
apply. Therefore, the Roll-Stack Stiffness Model and Frame Stiffness Model did not use initial
conditions. The Lumped Mass Model could have made use of initial conditions as it is a dynamic
model; however, initial conditions were not used as the rotation of rolls and the translation of the
strip were not simulated. The 2-D and 3-D Rolling Models made use of initial conditions, which
allowed them to start with the strip and rolls already at steady-state speeds. The initial conditions
removed the need to simulate the strip and rolls reaching steady-state speeds, the simulation sim-
ply started with those conditions. Figure 3.7 shows the initial conditions for the Rolling Model.
3.2.8 Constraints
Constraints are similar to boundary conditions in that they describe how nodes are constrained
to move relative to each other. Constraints are used in two places in the nite element models.
29
Work Roll
Backup Roll
Backup Roll
Work Roll
Backup Roll
Strip
Roll-Stack Stiffness Model
Frame Stiffness Model
Strip Rolling Model
Work Roll
Symmetry BC
Encastre BC
Horizontal Fixed BC
Vertical Fixed BC
Bending Load BC
F
r
a
m
e
F
r
a
m
e
Velocity BC
Displaced Symmetry BC
Figure 3.6: Model boundary conditions.
Axis of Rotation
Velocity Direction
Velocity IC
Figure 3.7: Rolling Model initial conditiions.
30
Chapter 3. Rolling Model Development 3.3. Strip Model
The rst place is to tie the shell of the roll to the core. This tie constraint requires that the nodes
on the inside surface of the shell cannot separate from the nodes on the outside surface of the
core, where the nodes touch. The second instance of constraints occurs in the 3-D nite element
models where the degrees of freedom of the bearing surface were tied to the degrees of freedom
of a single reference point. This allows load and boundary conditions to be applied at one node
and affect an entire surface. This technique was used at the end of the roll necks in the 3-D nite
element models to restrict their motion and apply bending forces.
3.2.9 Implicit and Explicit Methods
Dynamic nite element models must integrate the equations of motion to represent inertial effects.
This integration is performed using either explicit or implicit integration methods to represent
time dependent problems. The difference between the two methods lies in how they integrate the
equations of motion. Explicit methods formulate the equation of motion such that the solution at a
time step is only dependent on the conditions in the previous step. This method has the advantage
of being very efcient as iteration is not needed. However, the solution is highly dependent on the
time step, which can cause instability if its too large. The solution of an implicit method at a given
time step is not only dependent on the conditions in the previous step, but also dependent on
unknowns in the current step. For this reason, the method must use iteration to nd the solution
at the current step. While implicit methods do require iteration, they are unconditionally stable.
Therefore, the choice of time step does not affect the solution [30].
The static problem of the roll-stack and frame stiffness models was solved using the implicit
method. Since the purpose of those models was to obtain a load-displacement curve with no time-
dependent effects, the implicit method was the most effective solver because it is unconditionally
stable and allows the step time to be controlled to give a specic number of points in the curve.
The dynamic models used the explicit method to obtain time-dependent solutions of the rolling
problem. The explicit method was able to solve the highly nonlinear contact between the strip and
rolls where the implicit model would have had difculties due to mesh deformation. In addition,
the large strains seen in the deformation of the strip were easily handled by the explicit method.
Lastly, the explicit method is computationally efcient which reduced the run time of the large
3-D rolling model [30].
3.2.10 Finite Element Analysis Results
After solving for nodal displacements via the integration methods, information can be extracted
and analyzed from the results. Many types of information are returned from nite element anal-
ysis such as element stresses and strains along with contact pressures and reaction forces. These
quantities provide the basis for the analysis of the rolling process.
3.3 Strip Model
The strip model consists of a strip material model and a simplied stiffness model. The strip used
an elastic-plastic model for the nite element analysis performed in this work. The strip stiffness
model was based upon the rolling load model of Ford and Alexander [4].
3.3.1 Material Model
Information on the material properties of the strip as it passes through the stands is difcult to ob-
tain. The test data provided a predicted yield stress based on the online controllers reaction to the
31
material. The mill controllers use the Ford and Alexander rolling load model for this prediction.
The material model used in the nite element models was based upon an elastic-plastic material.
Since the nal model will be applied to materials that may not be steel, such as aluminum, it
was not reasonable to limit it to one material model. Ford and Alexander assumes a rigid plastic
material which has no strain hardening effects.
The following curves are constructed in terms of true stress and true strain. True strain
T
is
dened in Equation 3.4
t
= ln(1 +) (3.4)
where is engineering strain. The true stress
t
is based on engineering stress and engineering
strain as shown in Equation 3.5.
t
= (1 + ) (3.5)
An elastic-plastic material model was used for the construction of the nite element rolling
models. A graphical representation of the stress-strain curve for the material model is shown in
Figure 3.7. The elastic region was dened by the elastic modulus E of the material. For the models
in this work, the elastic modulus was assumed to be 18 tonne/mm
2
, slightly less than the actual
elastic modulus of steel because of temperature effects. The yield stress S
y
was given by the test
data as 0.02118574 tonne/mm
2
. The initial yield strain
1
was found by Equation 3.6.
1
=
S
y
E
(3.6)
The end point of the stress-strain curve
2
was set to 100% strain as the estimated strain of the
element was less. The yield stress coefcient was used to provide the value of the true stress at
100% true strain. This variable was originally going to be used in a material calibration, but it was
found to have little effect on the rolling load and exit thickness of the strip. A value of two was
chosen for the yield stress coefcient such that the stress at the end of the stress-strain curve was
2S
y
.
Figure 3.8: Elastic-plastic stress-strain curve with elliptical strain hardening.
An elliptical curve was used to dene the plastic region of the stress-strain curve as shown in
Figure 3.7. The shape of the ellipse is dened in Equation 3.7 which can then be solved for as in
Equation 3.8.
32
(
2
)
2
(
2
1
)
2
+
( S
y
)
2
(S
y
)
2
= 1 (3.7)
=
_
1
(
2
)
2
(
2
1
)
2
_
(S
y
)
2
(3.8)
With these parameters dened, the stress-strain curve for the nite element models was generated
as shown in Figure 3.9. The elastic region of the model appears vertical due to the large strains
the model covers. These stress-strain curves represent the true-stress true-strain condition of the
material as opposed to the engineering-stress engineering-strain.
Figure 3.9: Strip stress-strain curve used in the FE rolling models.
3.3.2 Strip Stiffness
For the lumped-mass model in section 3.7 the reaction force of the strip must be represented
as a spring. Ginzburg provided one method of accomplishing this task as detailed in section
2.2.1. However, Ginzburgs model is a special case of the generalized method used in the lumped-
mass model. Since the lumped-mass model represents the actions of the roll-stack at steady-state,
determining how changes in exit thickness of the strip affect the rolling load can give a pseudo
stiffness of the strip. The strip exit modulus k
x
is dened by the industry as
k
x
=
P/w
H
X
(3.9)
33
Chapter 3. Rolling Model Development 3.4. Model Simplications
where P/w is the rolling load per unit width of the strip and H
X
is the exit thickness of the strip.
While this derivative cannot be taken analytically because there is no closed form solution to the
rolling load equations, it is possible to nd it through nite differencing. The Python code in Ap-
pendix: Rolling Load Model was used to calculate the rolling load based on the stand parameters.
The exit thickness is perturbed above and below the predicted exit thickness of 5.497 mm by 2.5%.
The rolling load is then calculated for both cases. The partial of rolling load is then found with
respect to exit thickness by Equation 3.10.
k
x
=
P/w
H
X
=
(P
1
P
2
)/w
H
X
1 H
X2
(3.10)
Performing this calculation based on the test data, gave a k
x
of 0.608 tonne/mm
2
. Then, multiply-
ing by the width of the strip provides the strip stiffness k
s
as shown in Equation 3.11
k
s
= k
x
w (3.11)
which gave a k
p
of 927.8 tonne/mm for a strip 1589.5 mm wide. This value approximates the
stiffness of the strip at steady-state which was then used in the lumped-mass model.
3.4 Model Simplications
To reduce the complexity of the models in this project, the following assumptions were made:
stand symmetry can reduce the amount of the stand that must be modeled, the stand frame can
be represented by a spring, the roll bearings hold the neck such that it cannot rotate in bending,
and the draft of the strip can be modeled by displacing the strip into the rolls.
3.4.1 Stand Symmetry
Figure 3.10 shows the symmetry of the stand. With this assumption the entire stand was repre-
sented by a quarter model when the rolls were rotating, or by an eighth model when the rolls were
not rotating. The roll-stack stiffness and frame stiffness models do not require the rolls to rotate
and so where modeled by using an eighth stand representation. The 2-D and 3-D rolling models
required the rolls to rotate and so used a quarter stand representation
3.4.2 Frame Stiffness
The stands roll-stack and frame were modeled as a series of springs as shown in Figure 3.11. By
inspecting the stand, it was possible to reduce the roll-stack to a stiffness k
r
that was in series with
the frame of stiffness k
f
. The total stiffness of the system k
t
was related to k
f
and k
r
by Equation
3.12.
k
t
=
_
1
k
r
+
1
k
f
_
1
(3.12)
The total stiffness k
t
was determined by the stand calibration performed at the mill. The roll-stack
stiffness k
r
was determined by a static implicit FE model. Thus, it was possible to solve for the
frame stiffness k
f
as in Equation 3.13.
k
f
=
_
1
k
t
1
k
r
_
1
(3.13)
34
Figure 3.10: Stand Symmetry, front and side views respectively.
Roll Stack
Stand
Gap Actuator
(a) (c) (d) (e) (b)
Figure 3.11: Stand stiffness representation. (a) Stand. (b) Spring representation of stand & roll stack. (c)
Simplication of b. (d) Roll-stack and stand spring in series. (e) Total stiffness of system.
35
This method was used by the roll-stack model to calculate the frame stiffness k
f
. The total stiffness
of the roll-stack and frame was veried by the frame stiffness model using Equation 3.12. The
value k
f
is used as a lumped-parameter stiffness of the spring in the FE models which represent
stand stiffness.
3.4.3 Strip Displacement Curve
To shorten run times, it was essential that the strip and rolls be brought into contact in a manner
that minimized impact loads. Instead of setting the rolls a given distance apart and letting the strip
impact the gap, the strip was brought into contact with the rolls by displacement. The process is
illustrated in Figure 3.12. When the simulation starts, the roll and strip are aligned and in contact.
As the simulation runs, the centerline of the strip is displaced upward. The displacement distance
d
t
is calculated in Equation 3.14
d
t
=
H
E
H
g
2
(3.14)
where H
E
is the entry thickness of the strip and H
g
is the gap between the work rolls. The divisor
in the equation is due to half of the strip being modeled.
Strip Centerline Displacement
A B
Figure 3.12: Strip displacement. (A) Initial system setup. (B) Location of the strip after displacement.
The displacement curve was used to describe the transition from Figure 3.12A to Figure 3.12B
by specifying the location of the strip centerline over time. A double harmonic motion curve was
used to describe the displacement as given in Equation 3.15
d =
d
t
2
__
1 cos
t
t
_
1
4
_
1 cos
2t
t
__
(3.15)
where d is the displacement at time t,
t
is the total time in which the displacement occurs, and d
t
is the displacement [25]. The displacement curve is shown in Figure 3.13.
36
Chapter 3. Rolling Model Development 3.5. Static Roll-Stack Stiffness Model
Figure 3.13: Strip displacement curve.
3.5 Static Roll-Stack Stiffness Model
The Static Roll-Stack Stiffness Model was the rst nite element model used to determine data
necessary to simulate the rolling process. The roll-stack in the model refers to the collection of
the four rolls in the four-high roll-stack stand conguration. The model is considered static as
the dynamic response of the roll stack was not included. The eighth roll-stack nite element
model was implemented in both 2-Dand 3-Das the 2-Dand 3-Dmodels provide slightly different
stiffnesses. The variation in stiffness is due the assumption of plane strain in the 2-D model and
its inability to represent the roll geometry, such as the neck, which the 3-D model can represent.
3.5.1 Purpose
The purpose of the Roll-Stack Stiffness Model was to determine the stiffness of the roll stack. This
information is necessary as it allows the stiffness of the frame to be found given the total frame
and roll-stack stiffness from the test data. The roll-stack stiffness was determined by modeling
one eighth of the total roll stack, a simplication that was justied due to the symmetry of the
stand. Figure 3.14 shows the calculation of the roll stack stiffness k
r
from the eighth model. Due
to symmetry conditions, the stiffness of the eighth model is one half that of the total roll-stack.
3.5.2 Procedure
In the roll-stack model, the backup roll bearing was held xed and the stack was compressed by
displacing the bottom symmetry condition of the work roll upward as shown in Figure 3.14. The
balance bending force was applied to the work roll bearing throughout the process. The boundary
37
Figure 3.14: Eighth model of the roll stack to determine stack stiffness
was displaced such that the maximumreaction force at the backup roll bearing was approximately
1500 tonnes.
Compressing the 2-D and 3-D roll stack models gave a load-deection curve from which stiff-
ness was determined. Taking the slope of this curve gave the stiffness of the entire roll stack,
which was used in Equation 3.13 to determine the frame stiffness k
f
. This analysis is shown in
section 3.5.4.
3.5.3 Model Setup
The Roll-Stack Stiffness Model was generated from information given in the test data. The rele-
vant test data for both the 2-D and 3-D models is presented in Table 3.1. The roll geometry and
material properties were generated from this data. Mesh sizes were set after performing a mesh
convergence study based on calculated stiffness. The mesh was considered sufciently converged
when the stiffness varied by less than 5% with changes in mesh size. Figure 3.15 shows the result-
ing partitioning and meshing scheme for the 2-D roll-stack models. The 3-D model used a similar
mesh as shown in Figure 3.16. Boundary conditions, discussed in Section 3.2.6, were applied as
shown in Figure 3.17. The 2-D model used reduced integration quadrilateral elements while the
3-D model used reduced integration hexahedral elements as discussed in Section 3.2.5.
Several boundary conditions were dened in the roll-stack model. The symmetry planes of the
rolls were constrained such that surfaces could move vertically, but not away from the planes of
symmetry. These constraints allowed the eighth model to simulate the full roll-stack. The backup
roll bearing was xed in place as shown in Figure 3.14. The work roll bearing was constrained
such that it could only move vertically.
3.5.4 Analysis & Results
The load-displacement data for the 2-D and 3-D models was extracted from the output databases
generated during the Abaqus solutions. This data was post-processed to account for symmetry
conditions to give the load-displacement curve of the entire roll-stack. Figure 3.18 and Figure 3.19
show graphs and curve ts for the load displacement data from the 2-D and 3-D models respec-
tively. In both data sets, the linear curve ts represent the deection of the roll-stack with little
error. From the ts, the stiffness of the roll stacks were extracted giving a k
r
of 1701.6 tonne/mm
for the 2-D model and a k
r
of 1865.55 tonne/mm for the 3-D model. These roll-stack stiffness k
r
38
Table 3.1: Roll-Stack Stiffness Model setup data.
Property Symbol Value Units Models
Work Roll Barrel Length B
lw
2080 mm 2-D & 3-D
Work Roll Base Diameter B
dw
655.715 mm 2-D & 3-D
Work Roll Neck Length N
lw
195 mm 3-D
Work Roll Neck Diameter N
dw
510 mm 3-D
Work Roll Barrel Dia Crown C
w
-0.16 mm 2-D & 3-D
Work Roll Shell Thickness t
sw
50 mm 2-D & 3-D
Backup Roll Barrel Length B
lb
1820 mm 2-D & 3-D
Backup Roll Base Diameter B
db
1589.505 mm 2-D & 3-D
Backup Roll Neck Length N
lb
360 mm 3-D
Backup Roll Neck Diameter N
db
945 mm 3-D
Backup Roll Barrel Dia Crown C
b
0 mm 2-D & 3-D
Backup Roll Shell Thickness t
sb
60 mm 2-D & 3-D
Roll Core Elastic Modulus E
cb
21.092 tonne/mm
2
2-D & 3-D
Roll Core Poissons ratio
cb
0.29 2-D & 3-D
Roll Shell Elastic Modulus E
sb
21.092 tonne/mm
2
2-D & 3-D
Roll Shell Poissons ratio
sb
0.29 2-D & 3-D
Balance Bending Force J
b
150 tonne 2-D & 3-D
Roll-Roll Friction
r
0.2 2-D & 3-D
Total Stiffness k
t
610 tonne/mm 2-D & 3-D
(a)
(b)
a
b
Figure 3.15: 2-D roll-stack model partition and meshing scheme.
39
Figure 3.16: 3-D roll-stack model partition and meshing scheme.
Work Roll
Backup Roll
Symmetry BC
Encastre BC
Horizontal Fixed BC
Vertical Fixed BC
Bending Load BC
Velocity BC
Figure 3.17: Roll-Stack Stiffness Model boundary conditions.
values were used with Equation 3.13 to predict frame stiffness k
f
values of 950.87 tonne/mm and
906.36 tonne/mm for the 2-D and 3-D models respectively given that the total stiffness k
t
was
610 tonne/mm.
3.5.5 Summary
The roll-stack model was based on a nite element modeling of one eighth of the stand. The roll-
stack model was used to calculate the stiffness of the roll-stack. The stiffness of the stack was then
used to predict the stiffness of the frame. Table 3.2 summarizes the data extracted from the 2-D
and 3-D roll-stack models and the predicted frame stiffnesses.
Table 3.2: Roll-stack model stiffness results and predictions.
Type k
t
[tonne/mm] k
r
[tonne/mm] k
s
[tonne/mm]
2-D 610.00 1701.62 950.87
3-D 610.00 1865.55 906.36
40
Figure 3.18: 2-D roll-stack load-displacement curve.
Figure 3.19: 3-D roll-stack load-displacement curve.
41
Chapter 3. Rolling Model Development 3.6. Static Frame Stiffness Model
3.6 Static Frame Stiffness Model
The Static Frame Stiffness Model differs from the Static Roll-Stack Stiffness Model only in bound-
ary conditions. 2-D and 3-D variations of the nite element model were constructed to calculate
the stiffness of the frame.
3.6.1 Purpose
The purpose of the Frame Stiffness model was to verify the predicted stiffness of the frame from
the Roll-Stack Stiffness Model. This verication was done by inserting a spring representing the
frame and loading the rolls. This method attempts to reproduce the stand calibration performed
in the mill which determines the total roll stack and frame stiffness. The model was constructed
as shown in Figure 3.20.
Figure 3.20: Eighth model of the roll stack to determine frame stiffness.
3.6.2 Procedure
The procedure for the Frame Stiffness Model is the same as for the Roll-Stack Stiffness Model
except for the distance the bottom symmetry boundary condition is displaced. In the stand cal-
ibration performed in the mill, the roll stack is loaded until the load sensor reads 1500 tonnes.
The same is applied in this model such that the boundary condition is displaced until the reaction
force is approximately 1500 tonnes.
3.6.3 Model Setup
All aspects of this model are the same as the roll-stack model except for the backup roll boundary
condition and the addition of the frame represented by a spring. The relevant test data is summa-
rized in Table 3.3. The mesh developed was the same as in the Roll-Stack Stiffness Model and may
be found in Figures 3.15 and 3.16. As discussed in Section 3.2.5, reduced integration quadrilateral
elements were used for the 2-D model and reduced integration hexahedrals were used for the 3-D
model. The boundary conditions from Section 3.2.6 were applied as shown in Figure 3.21.
42
Table 3.3: Static Frame Stiffness Model setup data.
Property Symbol Value Units Models
lw
2080 mm 2-D & 3-D
dw
655.715 mm 2-D & 3-D
lw
195 mm 3-D
dw
510 mm 3-D
Work Roll Barrel Dia Crown C
w
-0.16 mm 2-D & 3-D
sw
50 mm 2-D & 3-D
lb
1820 mm 2-D & 3-D
db
1589.505 mm 2-D & 3-D
lb
360 mm 3-D
db
945 mm 3-D
Backup Roll Barrel Dia Crown C
b
0 mm 2-D & 3-D
sb
60 mm 2-D & 3-D
Roll Core Elastic Modulus E
cb
21.092 tonne/mm
2
2-D & 3-D
Roll Core Poissons ratio
cb
0.29 2-D & 3-D
Roll Shell Elastic Modulus E
sb
21.092 tonne/mm
2
2-D & 3-D
Roll Shell Poissons ratio
sb
0.29 2-D & 3-D
b
150 tonne 2-D & 3-D
Roll-Roll Friction
r
0.2 2-D & 3-D
Total Stiffness k
t
610 tonne/mm 2-D & 3-D
2-D Frame Stiffness k
f
950.87 tonne/mm 2-D
f
906.36 tonne/mm 3-D
Backup Roll
Work Roll
Symmetry BC
Encastre BC
Horizontal Fixed BC
Vertical Fixed BC
Bending Load BC
F
r
a
m
e
Velocity BC
Figure 3.21: Roll-Stack Stiffness Model boundary conditions.
43
In the roll-stack model the backup roll bearing boundary condition was xed such that the
bearing could not move vertically. In this model the roll is able to move, though it is constrained
by the stand spring which connects the roll to ground. The frame stiffness is represented by a
spring element of stiffness k
f
equal to that predicted by the roll-stack model. In the 2-D model the
frame stiffness was 950.87 tonne/mm and in the 3-D model it was 906.36 tonne/mm.
The Frame Stiffness Model was run for both 2-D and 3-D cases. Load-displacement curves were
calculated from the reaction force at the frame encastre boundary condition and the displacement
of the symmetry boundary condition which compressed the rolls and frame. Figures 3.22 and 3.23
show the load-displacement curves for the 2-D and 3-D models respectively. The slopes of the
curves were extracted in the same manner as the roll-stack model and the predicted value of the
total stiffness was compared to the actual total stiffness of 610 tonne/mm. The 2-D case gave a
total stiffness of 611.46 tonne/mm and the 3-D case gave a total stiffness of 609.80 tonne/mm.
Figure 3.22: 2-D frame stiffness load-displacement curve.
3.6.5 Summary
The Frame Stiffness Model was used to verify the frame stiffness calculated from the data in
the Roll-Stack Stiffness Model. The analysis showed that the calculated frame stiffness yielded
a predicted total stiffness that was less than 1% different from the actual total stiffness. Table 3.4
summarizes these results. These values for the frame stiffness were used in all of the following
models.
44
Chapter 3. Rolling Model Development 3.7. Dynamic Lumped-Mass Model
Figure 3.23: 3-D frame stiffness load-displacement curve.
Table 3.4: Frame stiffness model results.
Type k
t
(Measured) [tonne/mm] k
t
(Predicted) [tonne/mm] k
t
% Error
2-D 610.00 611.46 0.24
3-D 610.00 609.80 0.03
3.7 Dynamic Lumped-Mass Model
The Dynamic Lumped-Mass Model is an explicit nite element model created to simulate the
dynamic reaction of the rolls oscillating in the stand. This model uses a two degree-of-freedom
lumped-mass representation of the four-high roll stack which is further simplied through stand
symmetry. The model most closely approximates the dynamics of the 3-D Rolling Model.
3.7.1 Purpose
There are two purposes for creating the Dynamic Lumped-Mass Model. The rst is to create a
model capable of simulating the dynamics of the roll stack. The second is determine methods
of controlling the dynamic response such that the roll stack reaches steady-state conditions as
quickly as possible thereby shortening the simulation run time for the rolling models. The top
quarter of the stand was modeled as a multiple-degree-of-freedom system as shown in Figure
3.24.
Since this model represents a quarter of the roll-stack and stand, m
w
and m
b
are half the mass
of the work roll and backup roll respectively. The frame stiffness is k
f
, the roll-stack stiffness is k
r
,
and the strip stiffness is k
s
. Dampers were included in the model to control the dynamic response.
45
1
2
3
4
Figure 3.24: Lumped mass model of a quarter roll stack.
The roll damping coefcient is c
r
and the frame damping coefcient is c
f
. The values for these
parameters are given in section 3.7.6.
3.7.2 Single Degree-of-Freedom System
The Lumped-Mass Model was developed to take advantage of explicit dynamic nite element
modeling techniques similar to those used in the later models, though much simplied due to
choice of representation. The foundation for the lumped-mass method is the single degree of
freedom mass-spring-damper system with no forcing function shown in Figure 3.25. The mass-
spring-damper system is represented by the equation of motion,
m x +c x +kx = F (3.16)
where m is mass, c is damping coefcient, k is stiffness, and F is a forcing function. The natural
frequency of the system is dened as
n
in Equation 3.17 [12].
n
=
_
k
m
(3.17)
A single degree-of-freedom system will oscillate indenitely without damping. There are
three conditions that can occur when damping is added depending on the value of the damp-
ing coefcient. The model can either be overdamped, critically damped, or underdamped. In
an overdamped system, there is no oscillation, but the system can require a long time to reach
steady-state. In a critically damped system there is no oscillation and the system goes directly
to steady-state usually in twice the period of the natural frequency. Lastly, underdamped sys-
tems oscillate, but the amplitude of the oscillation continually decreases until the system reaches
steady-state. These conditions are dened by the damping ratio given in Equation 3.18 [12].
46
Figure 3.25: Mass-spring-damper system [12].
=
c
2m
n
=
c
2
_
k/m
(3.18)
The damping coefcient can be greater than one (overdamped), equal to one (critically damped),
or less than one (underdamped). The responses of the single degree of freedom system with dif-
ferent damping ratios are shown in Figure 3.26. The solutions were computed using an ordinary
differential equation integration technique. It can be seen that the system with equal to one
reaches steady-state the quickest [12].
Figure 3.26: single degree of freedom system response.
47
3.7.3 Multiple Degree-of-Freedom System
Similar techniques can be applied to the multiple degree of freedom system of the roll-stack and
frame; however, the equations must be rewritten in matrix form. The equation of motion remains
the same, though values are replaced by mass, damping and stiffness matrices which represent the
coupling between the displacement degrees of freedom associated with each mass as in Equation
3.19.
M x +C x +Kx = F (3.19)
The matrices in the equation of motion are assembled by direct inspection. Equation 3.20 describes
the mass matrix of the system. Since there are four nodes, as shown in Figure 3.24, the matrix has
four rows and four columns because there are four degrees of freedom.
The nodes in the model are numbered from one to four in Figure 3.24. The rows and columns
in the mass, damping, and stiffness matrices correspond to values at those nodes. For instance,
row and column one reference node one.
M =
_
_
0 0 0 0
0 m
b
0 0
0 0 m
w
0
0 0 0 0
_
_
(3.20)
The stiffness and damping matrices, in Equations 3.21 and 3.22 respectively, are assembled by
inspection as well.
C =
_
_
c
f
c
f
0 0
c
f
c
f
+c
r
c
r
0
0 c
r
c
r
0
0 0 0 0
_
_
(3.21)
K =
_
_
k
f
k
f
0 0
k
f
k
f
+k
r
k
r
0
0 k
r
k
r
+k
s
k
s
0 0 k
s
k
s
_
_
(3.22)
Nodes one and four are xed, having a prescribed displacement of zero. Since these nodes cannot
move, their displacement needs not be solved for so the corresponding rows and columns can be
removed. This produces the reduced mass, damping, and stiffness matrices which represent the
system as given in Equation 3.23 [12].
M =
_
m
b
0
0 m
w
_
C =
_
c
f
+c
r
c
r
c
r
c
r
_
K =
_
k
f
+k
r
k
r
k
r
k
r
+k
s
_
(3.23)
The natural frequencies of the system were found by solving the eigenvalue problem represented
by Equation 3.24.
48
det
_
2
M +K
= 0 (3.24)
For the two degree-of-freedom system in Equation 3.23, the solution of Equation 3.24 resulted in
two natural frequencies,
1
and
2
. The in-phase vibration mode had a lower natural frequency
than the out-of-phase vibration mode [12].
As shown in the single-degree-of-freedom system, critically damping the multiple degree-
of-freedom system should cause the system to reach steady state in the least amount of time.
Critically damping a multiple degree-of-freedom system is more difcult because the interactions
between the masses may not decouple the contribution of the damping forces between the masses.
To ensure decoupling, proportional damping is commonly used. Proportional damping formu-
lates a damping matrix by combining the mass and stiffness matrices as shown in Equation 3.25
[12].
C
cr
= M +K (3.25)
The coefcients and are modal damping ratios related to the natural frequencies of the system
and the damping ratio by Equation 3.26 where
i
denotes each natural frequency of the system
and
i
is the damping ratio of each mode.
i
=

2
i
+

i
2
(3.26)
Given Equation 3.26 and the two degrees of freedom that the system is reduced to, it was possible
to solve for and in terms of the two natural frequencies
1
and
2
. Equations 3.27 and 3.28
dene and respectively [12].
=
2
1
1
+
2
(3.27)
=
2
1
+
2
(3.28)
With equal to one to critically damp the system, Equations 3.27 and 3.28 reduce to the values
shown in Equation 3.29.
=
2
1
1
+
2
=
2
1
+
2
(3.29)
Once and were solved for, the damping values c
r
and c
f
needed to critically damp the model
were found. Equation 3.30 shows how the systems mass, damping, and stiffness matrices can be
related through proportional damping.
C
cr
= M +K =
_
m
b
0
0 m
w
_
+
_
k
f
+k
r
k
r
k
r
k
r
+k
s
_
= (3.30)
However, the matrix that Equation 3.29 yields cannot be represented by the dampers in the
lumped-mass model of the physical roll stack stand model. This is because the matrices C and
C
cr
have different forms as shown in Equation 3.31.
C =
_
c
f
+c
r
c
r
c
r
c
r
_
C
cr
=
_
m
b
+(k
f
+k
r
) k
r
k
r
m
w
+(k
r
+k
s
)
_
(3.31)
49
Since the dampers in the model cannot exactly critically damp the model, an approximate solution
was used. Based on the matrices in Equation 3.31, expressions for c
f
and c
r
were found as shown
in Equation 3.32.
c
f
= m
b
+k
f
c
r
= k
r
(3.32)
These damping coefcients approximately critically damp the lumped-mass system.
3.7.4 Explicit Method Formulation
The explicit method as discussed in section 3.2 was used to calculate the time solution of the
lumped-mass model, specically the exit thickness H
X
of the strip over time. The explicit method
uses a differencing approach to approximate the velocity and acceleration of masses in the equa-
tion of motion as given by the second-order central difference Equations 3.33 and 3.34 [30]. The
variable u
n
is used to represent the displacement of a node at a given time step.
v
n
=
u
n+1
u
n1
2t
(3.33)
a
n
=
u
n+1
2u
n
+u
n1
(t)
2
(3.34)
Substituting u
n
, v
n
, and a
n
into Equation 3.19 yields Equation 3.35 and solving for u
n+1
gives
Equation 3.36 which represents the displacement of the nodes at the next time step [30].
M(u
n+1
2u
n
+u
n1
) +
t
2
C(u
n+1
u
n1
) + (t)
2
Ku
n
= (t)
2
F (3.35)
_
M +
t
2
C
_
u
n+1
= (t)
2
[F
n
Ku
n
] +
t
2
Cu
n1
+M(2u
n
u
n1
) (3.36)
3.7.5 Model Run Parameters
Parameters were set to dene the length of time to simulate with the model. If not enough time is
allowed, the model will not come to steady-state. If too much time is allowed, the users time is
wasted. The criteria used for the steady-state condition is determined by the exit thickness of the
strip varying by less than one percent. Many factors affect how long it will take for the system to
reach steady-state, such as the stiffness of the frame and strip and the mass of the work rolls. These
factors cannot be controlled as they are characteristics of the stand. However, the displacement
time and total run time are controllable.
Since the frequency of vibration of the rolls is dependent on the properties of the stand and
vary by stand, the model must be able to account for these differences. Thus, it is not possible to
tune the model for one stand as it must be extensible to stands of various sizes and properties.
To take this variation into account, the run time t
r
and displacement time t
d
were set based on
multiples of the period of oscillation as calculated from the multiple degree-of-freedom system
in section 3.7.3. The lowest natural frequency of the system was used to calculate the period of
oscillation T. These coefcients are denoted as
d
and
r
for displacement time and run time, re-
spectively. Expressions for displacement time and run time as functions of T are given in Equation
3.37
t
d
=
d
T t
r
=
r
T (3.37)
50
The displacement time coefcient,
d
directly affects the time the model requires to reach steady-
state. The coefcient
r
determines how long the model will run. Figure 3.27 shows the strip exit
thickness over time for different values of
d
. As
d
increases, the time required to reach steady-
state increases. However, lower values of
d
causes the amplitude of oscillation to increase. If
the amplitude is too large, it could cause problems for the nite element models. Figure 3.28
displays rolling load versus time. Decreasing
d
has a similar effect on the amplitude of the
rolling load, causing it to peak much higher than the steady-state value. The run-time coefcient
r
is necessary for the nite element models as they do not yet use Abaquss internal approaches
to test for steady-state response. If
r
is too large, the model will take longer than necessary to
run; however, if
r
is too small, the model will not come to steady-state. The lumped-mass model
was used to estimate
r
and
d
for use in the rolling models.
Figure 3.27: Strip exit thickness H
X
versus time t.
3.7.6 Lumped-Mass Model Implementation
Equation 3.36 was implemented using the Python scripting language to solve for the time-domain
dynamic response of the lumped-mass roll-stack explicit model. Spring and damper elements
were used as shown in Figure 3.29. The modied system of the explicit model is slightly different
from the multiple degree-of-freedom system of Section 3.7.3 which was used to nd the natural
frequency and damping coefcients. The values for the parameters used in the Lumped-Mass
Model are shown in table 3.5.
The spring representing the strip was replaced by the rolling load P which was approximated
at each time step using the Ford and Alexander rolling load model. This substitution better rep-
resents the nonlinear load generated by rolling the strip than a spring of stiffness k
s
which is only
51
Figure 3.28: Rolling load P versus time t.
Figure 3.29: Lumped-mass explicit model.
52
Table 3.5: Lumped-Mass Model setup data.
Property Symbol Value Units
Work Roll Mass m
w
0.0003112 tonne-s
2
/mm
Backup Roll Mass m
b
0.0016382 tonne-s
2
/mm
Frame Stiffness k
f
927.21 tonne/mm
Roll-Stack Stiffness k
r
1865.55 tonne/mm
Strip Stiffness k
s
906.36 tonne/mm
Roll Damping Coefcient c
r
0.9200 tonne-s/mm
Frame Damping Coefcient c
f
2.7825 tonne-s/mm
Displacement-Time Coefcient
d
0.875
Run-Time Coefcient
r
2.25
Displacement Time t
d
0.005958 s
Run Time t
r
0.015320 s
valid near the steady-state. The mass, damping, and stiffness matrices used in the explicit model
are given in Equation 3.38.
M =
_
m
b
0
0 m
w
_
C =
_
c
f
+c
r
c
r
c
r
c
r
_
K =
_
k
f
+k
r
k
r
k
r
k
r
_
(3.38)
The rolling load P is calculated at each time step based on the strip exit thickness, and the stand
parameters. The exit thickness H
X
of the strip at each step is assumed to be equal to the gap
H
g
which varies according to the displacement curve and work roll displacement u
2
. Since the
displacement time t
d
represents the time in which the strip is displaced,
t
from Equation 3.15 can
be replaced by t
d
to form Equation 3.39
d =
d
t
2
__
1 cos
t
t
d
_
1
4
_
1 cos
2t
t
d
__
(3.39)
where d is the displacement of the strip centerline at time t and d
t
is the total displacement. Thus,
the exit thickness, H
X
is given by Equation 3.40.
H
X
= H
E
+u
2
2d (3.40)
This exit thickness, along with the work roll and strip properties can be used to predict the rolling
load P at each time step. This rolling load is then applied to the work roll at the next time step.
Eventually the rolling load and work roll displacement reaches steady-state.
3.7.7 Results
Figures 3.30 and 3.31 show the rolling load and exit thickness versus time from the lumped-mass
model. They verify that the model comes to steady-state in 2.25 periods with a displacement time
of 0.875 periods criteria given for the steady-state condition in section 3.7.5. Table 3.6 shows the
natural frequencies of the system, the oscillation period, maximum and steady-state rolling loads
and exit heights. These results were used in the construction of the 2-Dand 3-DFE rolling models.
53
Figure 3.30: Lumped-mass model strip exit thickness H
X
versus time t.
Figure 3.31: Rolling load P versus time t.
54
Chapter 3. Rolling Model Development 3.8. Dynamic 2-D Rolling Model
Table 3.6: Lumped-mass model results.
Description Value
First Natural Frequency 146.86 Hz
Second Natural Frequency 498.57 Hz
Oscillation Period 0.0068089 s
Maximum Rolling Load 1373.12 tonne
Steady-State Rolling Load 1130.73 tonne
Minimum Exit Height 5.2384 mm
Steady-State Exit Height 5.5010 mm
Strip Centerline Displacement 1.4726 mm
Time to Steady-State 0.01275 s
3.8 Dynamic 2-D Rolling Model
The Dynamic 2-D Rolling Model was created to ll the gap between the Dynamic Lumped-Mass
Model and the Dynamic 3-D Rolling Model. The 2-D Rolling Model provided a quick and simple
method to test methods of simulating the rolling process in Abaqus without the added complexity
of the 3-D Rolling Model.
3.8.1 Purpose
The purpose of the dynamic 2-D nite element model was to validate the results of the Lumped-
Mass Model and work with the partitioning and meshing before moving on to the 3-D nite
element model. The rotation of the rolls and motion of the strip is described in Figure 3.32
Core
Shell
Strip
Ground
Bearing
Spring
Damper
Figure 3.32: 2-D rolling model detail.
55
3.8.2 Procedure
The model was constructed based on the test data in Table 3.8 and as described in Section 3.8.3.
The model was generated in Abaqus using the Python scripting language. The natural frequency
of the rolls was approximated as in Section 3.7.5 with the Lumped-Mass Model. The same run-
time and displacement-time coefcients,
r
and
d
respectively, as in the Lumped-Mass Model
were used to calculate the displacement time and run time. The natural frequencies were also
used to calculate the roll and frame damping coefcients. The values of these parameters are
summarized in Table 3.7.
The model was used to generate an input le that Abaqus used to run the simulation. The
state of the system was taken at 100 equally spaced time intervals throughout the solution time
and all information is recorded into a database le. This data includes stress, strain, reaction
forces, displacements, and deformed shapes. The deformed thickness of the strip and the reaction
force at the ground node were of primary interest. Running the simulation required less than ve
minutes.
Table 3.7: 2-D Rolling Model run data.
Period of Oscillation T 0.00681 s
Natural Frequency 1 f
1
146.87 Hz
2
498.57 Hz
Run-Time Coefcient
r
2.25
d
0.875
Run Time t
r
0.01532 s
Displacement Time t
d
0.00596 s
r
0.8587 tonne-s/mm
f
2.816 tonne-s/mm
3.8.3 Model Setup
The 2-DRolling Model was generated fromthe test data and parameters in Tables 3.8 and 3.7. The
roll diameters and shell thicknesses were used to construct the rolls. The run time t
r
, calculated
from the run-time coefcient and the systems lowest natural frequency, was used to nd how far
the work roll would rotate given its angular velocity. The strip length was dened by the length
of the arc through which the work roll would rotate from Equation 3.41.
l
s
= Rt
r
+L
p
(3.41)
The length of the strip is l
s
, R is the radius of the work roll, t
r
is the run time,
w
is the angular
velocity of the roll in radians per second, and L
p
is the estimated arc of contact from Ford and
Alexander. This calculation gave a strip length of 109.86 mmfor a simulation run time of 0.01532 s.
The partitions on the roll shells were then set so that the nely meshed area was slightly longer
than the length of the strip. The resulting partitions and mesh are shown in Figure 3.33.
Boundary conditions for the 2-D Rolling Model are shown in Figure 3.34. The centers of the
rolls were xed such that they could move vertically but not horizontally. The strip was given a
displacement symmetry boundary condition to move the strip into place via the strip displace-
56
Table 3.8: 2-D Rolling Model setup data.
Strip Entry Thickness H
E
6.4337 mm
Strip Elastic Modulus E
s
18 tonne/mm
2
Strip Yield Stress S
y
0.021186 tonne/mm
2
Strip Poissons Ratio
s
0.3
Strip Density
s
7.9596e-13 tonne-s
2
/mm
4
Strip Velocity v 6752.64 mm/s
Strip Length l
s
109.86 mm
dw
655.715 mm
sw
50 mm
Work Roll Core Elastic Modulus E
cw
21.092 tonne/mm
2
Work Roll Core Poissons ratio
cw
0.29
Work Roll Shell Elastic Modulus E
sw
21.092 tonne/mm
2
Work Roll Shell Poissons ratio
sw
0.29
Work Roll Density
w
7.9596e-13 tonne-s
2
/mm
4
Work Roll Angular Velocity
w
20.596 rad/s
db
1589.505 mm
sb
60 mm
Backup Roll Core Elastic Modulus E
cb
21.092 tonne/mm
2
Backup Roll Core Poissons ratio
cb
0.29
Backup Roll Shell Elastic Modulus E
sb
21.092 tonne/mm
2
Backup Roll Shell Poissons ratio
sb
0.29
Backup Roll Density
b
7.9596e-13 tonne-s
2
/mm
4
Backup Roll Angular Velocity
b
-8.497 rad/s
Stand Bending Force J 96.88 tonne
Stand Gap H
g
3.9831 mm
Stand Strip-Roll Friction
s
0.2
Stand Roll-Roll Friction
r
0.2
Stand Total Stiffness k
t
610 tonne/mm
2-D Roll-Stack Stiffness k
r
1701.62 tonne/mm
f
950.87 tonne/mm
Strip Stiffness k
s
927.21 tonne/mm
57
(b)
(a)
a
b
Figure 3.33: 2-D rolling model mesh.
ment curve discussed in Section 3.4.3. An area of the cross section of the roll representing the
neck was given a rotational boundary condition which drove the work roll. The bending load
was applied to the center of the work roll effectively reducing the rolling load. The backup roll
center was connected to the frame spring which in turn was connected to ground with an encastre
boundary condition.
Symmetry BC
Encastre BC
Horizontal Fixed BC
Vertical Fixed BC
Bending Load BC
F
r
a
m
e
Velocity BC
Figure 3.34: 2-D Rolling Model boundary conditions.
Initial conditions were prescribed for the 2-D Rolling Model. The initial conditions allow the
simulation to start with the rolls rotating and the strip translating at approximately steady-state
velocities. The work roll rotates at 20.60 rad/s counterclockwise while the backup roll rotates at
8.50 rad/s clockwise. These values were calculated from roll diameters based on the test data
indicating that the strip was moving at 6752.64 mm/s. Figure 3.35 shows the application of these
initial conditions to the model.
58
Roll
Axis
Direction
Figure 3.35: 2-D Rolling Model initial conditions.
There are two regions in the 2-DRolling Model where contact between bodies occurs: between
the work roll and strip, and between the work roll and backup roll. A coefcient of friction of 0.2
was used in both contact regions. Hard contact was used with a penalty constrain method. A
coefcient between 0.2 and 0.3 is typical when the rolls are lubricated [24].
The rolling load and the exit thickness of the strip were extracted from the output database for
post-processing. These two outputs were used to determine when the system reached steady-
state. The system must reach steady-state before the exit thickness and rolling load can be mea-
sured and compared with the test data. The steady-state region was selected from a window of
15% of the data. This window was moved along the data and an error variance within the win-
dow was calculated at each position. The window with the least error variance was chosen as the
steady-state region. Figure 3.36 shows the rolling load versus time for the simulation. The red
line denotes the steady-state region of the simulation. The steady-state rolling load was found by
averaging the rolling load in the steady-state region which gives a rolling load of 976.28 tonnes.
Figure 3.37 shows how the exit thickness data is extracted from the model. The prole of the
strip, represented by the red line in the Figure, is used to determine the exit thickness along the
length of the strip. This exit thickness along the strip length is then plotted in Figure 3.38. From
this graph, it is possible to determine when the strip reaches steady-state in the same manner as
the rolling load. Again, the exit height is averaged in the stead-state region to give an average exit
thickness of 5.6129 mm.
3.8.5 Summary
The predicted rolling load of 976.28 tonnes differs from the test data rolling load of 1107.20 tonnes
by 11.82%. However, the exit thickness differs from the test data value of 5.4969 mm by 2.11%
with a predicted value of 5.6129 mm. The simulation steady-state region is from approximately
0.008 to 0.012 s which is similar to the results for the Lumped-Mass Model at 0.01275 s. These
results are summarized in Table 3.9. The later time on the Lumped-Mass Model was likely due
to a more stringent steady-state requirement. The response in rolling load and exit thickness was
59
Steady State
Figure 3.36: 2-D simulation rolling load P versus time t.
Figure 3.37: Strip prole for determination of exit thickness.
60
Steady State
Back - Strip - Front
Figure 3.38: 2-D exit thickness versus strip length.
also smooth enough that the meshing scheme appears to have been sufciently rened to give a
reasonable solution. With the strip exit height as the metric, the mesh was converged to within
ve percent.
The 2-Drolling model provided another step towards developing a complete roll-stack model.
While the 2-D model cannot predict crown, it simulates the system dynamics and shows that the
methods developed in the Lumped-Mass Model can be used to control and predict the response
of the system. The next step was to develop the 3-D Rolling Model capable of representing the
entire stand and predicted the strip exit crown.
Table 3.9: 2-D Rolling Model results summary.
Description Variable Unit Predicted Actual % Error
Rolling Load P tonne 976.28 1107.2 11.82
Exit Thickness H
X
mm 5.6129 5.4969 2.11
3.9 Dynamic 3-D Rolling Model
The dynamic 3-D nite element model is the next step beyond the 2-D rolling model. The 3-D
rolling model is the goal of this work. It is able to represent the rolls and strip as described in the
test data. The model is able to measure the strip crown exiting the stand and predict the rolling
loads and exit thickness accordingly. Many aspects of the 3-D Rolling Model are similar to the 2-D
Rolling Model.
61
3.9.1 Purpose
The purpose of the 3-D model is the same as the purpose of this work, to create a dynamic nite
element model capable of predicting strip prole. The model is able to represent the geometry
of the rolls and the material properties of the strip such that by simulating the rolling strip, it is
possible to predict the rolling load, exit thickness, and crown of the strip.
3.9.2 Procedure
The 3-D Rolling Model was constructed from the parameters given in the test data. In addition,
the run-time and displacement-time coefcients used in the Lumped-Mass Model and 2-DRolling
Model were also used in this model. Since the Lumped-Mass Model, 2-D Rolling Model, and 3-D
Rolling Model all represent the same geometry, the same natural frequencies, damping coefcients
and run times were found. These run-time parameters are summarized in Table 3.10.
Table 3.10: 3-D Rolling Model run data.
Strip Stiffness k
s
924.38 tonne/mm
r
1865.55 tonne/mm
f
906.36 tonne/mm
1
146.79 Hz
2
498.36 Hz
Run-Time Coefcient
r
2.25
d
0.875
Run Time t
r
0.01532 s
Displacement Time t
d
0.00596 s
r
0.9204 tonne-s/mm
f
2.7816 tonne-s/mm
Boundary conditions and initial conditions were then applied as discussed in Section 3.9.3. The
model was used to generate a simulation which was run in Abaqus. Running the simulation of
the 3-D Rolling Model required signicantly more time than the 2-D Rolling Model. The increase
in run time is due to the increase in the number of degrees of freedom in the model. Moving
from 2-D to 3-D in nite element analysis often causes a large increase in computation time. The
simulation ran to completion in approximately 55 minutes. The state of the system was captured
every 1.5e-4 s, 100 times during the simulation. These states were stored in an output database
and included data such as stresses, strains, displacements, deformations, and reaction forces. This
data provides the information necessary to calculate the rolling load, exit thickness, and strip
crown which the model predicts.
3.9.3 Model Setup
The geometry of the rolls was created from the test data described in Table 3.11. The roll crown
was made by drawing a spline through points to create a curve, which was then revolved to
make the roll barrels. The strip prole was dened by a spline from the center of the strip to its
62
edge. This prole was then extruded to form the strip. The length of the extrusion and length
of the strip was calculated in the same manner as in Section 3.8.3. The partitions were also sized
as described in the 2-D rolling model such that only the nely meshed regions were in contact
during the simulation. The mesh is shown in Figure 3.39.
Figure 3.39: 3-D rolling model mesh.
Figure 3.40 illustrates the boundary conditions for the 3-D Rolling Model. The centers of the
rolls were xed such that they could move vertically but not horizontally. The strip was given a
displacement symmetry boundary condition to move the strip into place via the strip displace-
ment curve discussed in Section 3.4.3. The end surface of the work roll neck was given a rota-
tional boundary condition which simulates the motor driving the work roll. The bending load
was applied to the same surface as the rotational boundary condition. The backup roll center was
connected to the frame spring which in turn was connected to ground with an encastre boundary
condition. A symmetry boundary condition was applied at the middle of the rolls and strip along
their widths.
Work Roll
Backup Roll
Strip
Symmetry BC
Encastre BC
Horizontal Fixed BC
Vertical Fixed BC
Bending Load BC
F
r
a
m
e
Velocity BC
Figure 3.40: 3-D Rolling Model boundary conditions.
63
Table 3.11: 3-D Rolling Model setup data.
Strip Width w 1524.84 mm
Strip Feather w
f
40 mm
Strip Entry Thickness H
E
6.4337 mm
Strip Elastic Modulus E
s
18 tonne/mm
2
Strip Yield Stress S
y
0.021186 tonne/mm
2
Strip Poissons Ratio
s
0.3
Strip Density
s
7.9596e-13 tonne-s
2
/mm
4
Strip Velocity v 6752.64 mm/s
Strip Length l
s
110.83 mm
dw
655.715 mm
lw
2080 mm
lw
195 mm
dw
510 mm
Work Roll Barrel Diameter Crown C
w
-0.16 mm
sw
50 mm
Work Roll Core Elastic Modulus E
cw
21.092 tonne/mm
2
Work Roll Core Poissons ratio
cw
0.29
Work Roll Shell Elastic Modulus E
sw
21.092 tonne/mm
2
Work Roll Shell Poissons ratio
sw
0.29
Work Roll Density
w
7.9596e-13 tonne-s
2
/mm
4
Work Roll Angular Velocity
w
20.5963 rad/s
db
1589.505 mm
lb
1820 mm
lb
360 mm
db
945 mm
Backup Roll Barrel Diameter Crown C
b
0 mm
sb
60 mm
Backup Roll Core Elastic Modulus E
cb
21.092 tonne/mm
2
Backup Roll Core Poissons ratio
cb
0.29
Backup Roll Shell Elastic Modulus E
sb
21.092 tonne/mm
2
Backup Roll Shell Poissons ratio
sb
0.29
Backup Roll Density
b
7.9596e-13 tonne-s
2
/mm
4
Backup Roll Angular Velocity
b
-8.49653 rad/s
Stand Bending Force J 96.88 tonne
Stand Gap H
g
3.9831 mm
Stand Strip-Roll Friction
s
0.2
Stand Roll-Roll Friction
r
0.2
Stand Total Stiffness k
t
610 tonne/mm
64
The same initial conditions were prescribed for the 3-DRolling Model as the 2-Dmodel though
the geometry differed slightly. The initial conditions allow the simulation to start with the rolls
rotating and the strip translating at approximately steady-state velocities. The work roll rotates at
20.60 rad/s counterclockwise while the backup roll rotates at 8.50 rad/s clockwise. These values
were calculated from roll diameters based on the test data indicating that the strip was moving at
6752.64 mm/s. Figure 3.35 shows the application of these initial conditions to the model.
Roll
Axis
Direction
Figure 3.41: 3-D Rolling Model initial conditions.
There are two surfaces in the 3-D Rolling Model where contact between bodies occurs: be-
tween the work roll and strip, and between the work roll and backup roll. A coefcient of friction
of 0.2 was used in both contact regions. Hard contact was used with a penalty constrain method.
Data extraction in the 3-D model is slightly more complex than in the 2-D model as more infor-
mation is necessary to calculate crown. The strip exiting the rolls may be visualized as shown in
Figure 3.42. Only the right half of the strip was modeled, from the centerline to the edge because
the 3-D Rolling Model is a quarter symmetry model. The rolling load versus time, shown in Fig-
ure 3.43 was extracted from the reaction force in the frame, the same as the 2-D Rolling Model.
The exit thickness versus the strip length, Figure 3.44, was taken from the strip thickness along
the centerline, similar to the 2-D Rolling Model.
The steady-state rolling load extracted from Figure 3.43 was 936.70 tonnes, a 15.39% difference
from the test data rolling load of 1107.20 tonnes. The exit thickness predicted by the model was
5.7053 mm giving an error of 3.79% from the test data exit thickness of 5.4969 mm.
The crown is calculated fromthe steady-state region of the strip as selected in Figure 3.44. That
steady-state region is the same region represented in Figure 3.42. The hatching in the steady-state
region represents the elements. Each element has four nodes on the surface of the strip, one at
each corner. By taking all of the nodes in this region and plotting their thickness versus the strip
width, it is possible to create a graph of exit thickness versus strip width, represented by Figure
3.45. Because there will always be some variation in the strip along the length due to simulation
noise and discretization, the nodes will show some scatter. However, a curve can be t through
these points to smooth the data and make the crown calculation simpler. Since the assumed form
65
C
e
n
t
e
r
l
i
n
e
F
e
a
t
h
e
r F
r
o
n
t
E
d
g
e
S
t
e
a
d
y

S
t
a
t
e
R
e
g
i
o
n
Figure 3.42: 3-D rolling model strip detail.
Figure 3.43: 3-D rolling model load versus time.
66
Steady State
Back - Strip - Front
Figure 3.44: 3-D rolling model exit thickness versus strip length.
of crown is a parabola, the same form can be used for the t. Also, since the crown is only dened
fromthe strip centerline to the feather, the data points fromthe feather to the edge can be neglected
from the t.
Figure 3.46 shows the actual strip exit thickness versus width data generated by the model.
However, this Figure only covers the width from the centerline to the feather, not to the edge of
the strip, as the crown is calculated between the center and feather only. There are many more
data points across the width of the strip and in each cluster than in Figure 3.45; this is because
there are many more elements in the actual model. A parabola is t through the data and the
graph shows the t to have good correlation with the test data. The crown is calculated by taking
the thickness at the center of the parabola and subtracting the thickness at the feather. One could
also simply calculate the crown by taking the average value of the data points at the center during
steady state and subtracting from the average of the feather data points during steady state. The
crown calculated by the tting method is 0.1185 mm. The predicted crown has 275% error from
the actual crown of 0.0316 mm.
3.9.5 Mesh Convergence
A mesh convergence study was performed to determine how the discretization of the strip and
rolls affect the solution. Five simulations were run, each with a ner mesh in the contact region of
the rolls and strip. The number of elements in each simulation was recorded. The exit thickness
and strip prole were used as metrics to determine convergence. The exit thickness along the
strip is shown in Figure 3.47. The exit thickness versus strip width for each simulation is shown
in Figure 3.48. The simulations with the smallest number of elements had the noisiest results.
The exit thickness of the strip for each simulation was extracted from the data. The quality
67
E
x
i
t

H
e
i
g
h
t
Strip Half-Width
Strip Centerline Feather Edge
Crown Fit
Node Location
C
r
o
w
n
Figure 3.45: Sample representation of strip exit thickness versus width.
Figure 3.46: 3-D rolling model strip exit thickness versus width.
68
Sim 5
Sim 3
Sim 4
Sim 2
Sim 1
Figure 3.47: 3-D rolling model exit thickness convergence data.
Sim 1
Sim 2
Sim 5
Sim 3
Sim 5
Figure 3.48: 3-D rolling model crown prole convergence data.
69
of the parabolic curve t found based on the coefcient of determination R
2
. The number of
elements, exit thickness, and t quality are summarized in table 3.12.
Table 3.12: Mesh convergence data.
Sim Number of Elements Exit thickness R
2
1 31014 5.948 0.7718
2 33192 5.8851 0.8736
3 35832 5.7996 0.9480
4 40056 5.7476 0.9822
5 46128 5.7046 0.9990
Figure 3.49 shows exit thickness versus the number of elements in the simulation. The exit
thickness decreases with each renement in the mesh; however, the change in exit thickness be-
tween the last two points is less than one percent. Figure 3.50 shows the convergence of the t
quality R
2
. As the mesh is rened, there is less noise in the curve t. With respect to t qual-
ity there is less than two percent change in the last two simulations. These results show that the
simulation with 46128 elements is reasonably converged and is most likely not giving erroneous
results due to mesh discretization.
Figure 3.49: Mesh convergence, exit thickness versus number of elements.
70
Chapter 3. Rolling Model Development 3.10. Summary
Figure 3.50: Mesh convergence, t quality versus number of elements.
3.9.6 Summary
The 3-D Rolling Model was successfully created and was able to predict rolling load, exit thick-
ness, and strip crown. The initial results are shown in Table 3.13. A mesh convergence study was
performed to support the accuracy of these results in that the mesh was sufciently rened. The
next step with the 3-D Rolling model is to improve its accuracy, which will be dealt with in the
following chapter.
Table 3.13: 3-D Rolling Model results summary.
Exit Thickness H
X
mm 5.7053 5.4969 3.79
Exit Crown C
X
mm 0.1185 0.0316 275
3.10 Summary
Seven models were developed in the process of creating the nal 3-D rolling model. The roll-stack
model was used to determine the stiffness of the roll-stack. This data was then used to predict
the stand stiffness and the stand stiffness model was used to validate that prediction by matching
the calibration in the test data. The Lumped-Mass Model was used to explore the dynamics of the
system, indicating how long to make the runs and how quickly to displace the strip. The stand
stiffness from the rst two models and the run time parameters of the Lumped-Mass Model were
71
Chapter 3. Rolling Model Development 3.10. Summary
used in the 2-D rolling model to verify the system dynamics. Lessons in meshing and rolling
from the 2-D rolling model were then used to create the nal 3-D rolling model. The 3-D model
is capable of predicting the exit thickness, rolling load, and strip crown required by the project
goals.
The error in exit thickness, rolling load, and strip crown predicted by the 3-D Rolling Model
can be large, especially in the crown prediction. The next chapter documents two methods used
to reduce this error and then validates the calibrated model.
72
4 Calibration, Validation, & Sensitivity
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.2 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.3 Yield-Gap Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.4 Work Roll Crown Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.5 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.6 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.1 Introduction
The purpose of this chapter is to show how the stand parameters causing the error in rolling
load, exit thickness, and crown were determined and how they were used to reduce the error to
calibrate the model. Once the calibration was complete, the 3-D model was run again and nal
error estimates were found. Lastly, the model was used to determine strip crown sensitivity with
respect to controllable stand parameters.
4.2 Calibration
With the 3-D rolling model created, it became necessary to further calibrate the model before
generating the nal validation. These calibrations assumed that there were slight variations in the
test data that caused the prediction to not match the test data, especially with respect to crown.
All of the outputs of interest: exit thickness, rolling load, and crown, were very sensitive to slight
variations in model parameters. Thus, the result error could be reduced with slight adjustments
to the parameters.
4.2.1 Selection of Parameters
An examination of the stand parameters, from work roll barrel diameter to strip width, revealed
that most parameters were based on easily measured geometric properties or well known materi-
als. Only a few parameters could cause the resulting errors from measurement uncertainty.
Strip yield stress was one of these parameters. Small variations in yield stress caused large
changes in rolling load and exit thickness. The strip yield stress in the test data was measured by
the online controllers using the Ford and Alexander rolling-load model. This method assumed
that the rolling-load model accurately represented the rolling process, which may not be valid.
Variations in rolling load can be due to factors other than yield stress, entry thickness, and exit
thickness. For instance, changes in crown can cause a change in rolling load. In addition, the
representation of the strip material in the 3-D rolling model was different from that of the rolling-
load model, which may have been a source of error.
A second source of error was gap measurement. During the stand calibration at the mill, the
rolls were brought together until the load sensors read 1500 tonnes. The displacement of the rolls
73
Chapter 4. Calibration, Validation, & Sensitivity 4.3. Yield-Gap Calibration
was then set to zero. All gaps for the controllers were calculated from this point, so it is not
absolutely known when the rolls met. Thus, the gap measurement may be the source of some
error.
A third source of error was work roll thermal expansion. The test data was collected on a
strip that was rolled directly after a roll change operation. The work rolls in a mill are changed
every eight hours due to wear. Thus, the rolls were at room temperature when the strip entered
them. Initially it was assumed that the rolls would not be in contact with the strip long enough for
signicant heat transfer to occur. However, this assumption was reconsidered when the predicted
crown was signicantly different fromthe actual crown. Thermal expansion could cause the work
roll barrel diameter and crown to change. While the change in diameter would not signicantly
affect the model, a very small change in work roll crown could dramatically change the strip
crown.
Thus, the strip yield stress, gap, and work roll crown were selected as parameters which could
be varied to reduce the error in rolling load, exit thickness, and strip crown.
4.2.2 Calibration Method
Given that strip yield stress, gap, and work roll crown where chosen to calibrate the exit thick-
ness, rolling load, and exit crown, the question arose as to how to perform the calibration. Two
calibrations were used to minimize the three errors. By reexamining the parameters, it was found
that rolling load and exit thickness were mainly dependent on the yield stress and gap. Varia-
tions in work roll crown did not signicantly affect the rolling load and exit thickness. Thus, one
quadratic response surface was dened to nd the values of yield stress and gap that would min-
imize rolling-load and exit-thickness error. Then, a quadratic t was used to nd the work roll
crown that would minimize the strip crown error.
The rst calibration, termed the Yield-Gap Calibration, varied the strip yield stress and gap to
minimize the error in rolling load and exit thickness from the model prediction to the test data.
The second calibration, Work Roll Crown Calibration, varied the work roll crown to account for
thermal expansion to minimize strip crown error. Simulations were run after each calibration for
validation.
4.2.3 Model Setup
The information used to construct the models for the calibrations was the same as that used to
construct the 3-D Rolling Model in Section 3.9 except for the values that were varied during the
calibrations. Tables 4.1, 4.2, 4.3, 4.4, and 4.5 contain the properties and values necessary to con-
struct the models.
4.3 Yield-Gap Calibration
The yield-gap calibration seeks to minimize the error in rolling load and exit thickness by varying
the strip yield stress and gap.
4.3.1 Response Surface
Nine simulations were required to dene the quadratic response surface shown in Figure 4.1. The
surface is dened over an isoparametric space such that values of yield stress and gap may be
chosen arbitrarily. The interpolation functions at each node are dened in Equation 4.1 [22].
74
Table 4.1: Yield-Gap Calibration run data.
1
146.87 Hz
2
498.57 Hz
Run-Time Coefcient
r
2.25
d
0.875
Run Time t
r
0.01532 s
Displacement Time t
d
0.00596 s
r
0.8587 tonne-s/mm
f
2.816 tonne-s/mm
Table 4.2: Yield-Gap Calibration work roll data.
Base Diameter B
dw
655.715 mm
Barrel Length B
lw
2080 mm
Neck Length N
lw
195 mm
Neck Diameter N
dw
510 mm
Barrel Diameter Crown C
w
-0.16 mm
Shell Thickness t
sw
50 mm
Core Elastic Modulus E
cw
21.092 tonne/mm
2
Core Poissons ratio
cw
0.29
Shell Elastic Modulus E
sw
21.092 tonne/mm
2
Shell Poissons ratio
sw
0.29
Density
w
7.9596e-13 tonne-s
2
/mm
4
Angular Velocity
w
20.5963 rad/s
1
2
3
6
5 4
7
8
9
1 2 3
4 5 6
7
8 9
Figure 4.1: Isoparametric transformation of error space.
75
Table 4.3: Yield-Gap Calibration backup roll data.
Base Diameter B
db
1589.505 mm
Barrel Length B
lb
1820 mm
Neck Length N
lb
360 mm
Neck Diameter N
db
945 mm
Shell Thickness t
sb
60 mm
b
0 mm
cb
21.092 tonne/mm
2
Core Poissons ratio
cb
0.29
sb
21.092 tonne/mm
2
sb
0.29
Density
b
7.9596e-13 tonne-s
2
/mm
4
Angular Velocity
b
-8.49653 rad/s
Table 4.4: Yield-Gap Calibration strip data.
Width w 1524.84 mm
Feather w
f
40 mm
Entry Thickness H
E
6.4337 mm
Elastic Modulus E
s
18 tonne/mm
2
Yield Stress S
y
0.021186 tonne/mm
2
Poissons Ratio
s
0.3
Density
s
7.9596e-13 tonne-s
2
/mm
4
Velocity v 6752.64 mm/s
Length l
s
109.86 mm
Table 4.5: Yield-Gap Calibration stand data.
Bending Force J 96.88 tonne
Gap H
g
3.9831 mm
Strip-Roll Friction
s
0.2
Roll-Roll Friction
r
0.2
Total Stiffness k
t
610 tonne/mm
r
1865.55 tonne/mm
f
906.36 tonne/mm
76
1
=
1
4
(
2
)(
2
)
2
=
1
2
(1
2
)(
2
)
3
=
1
4
(
2
+)(
2
)
4
=
1
2
(
2
)(1
2
)
5
= (1
2
)(1
2
)
6
=
1
2
(
2
+)(
2
+)
7
=
1
4
(
2
)(
2
+)
8
=
1
2
(1
2
)(
2
+)
9
=
1
4
(
2
+)(
2
+)
(4.1)
The isoparametric representation allows yield stress S
y
, gap H
g
, and error , to be represented
parametrically as in Equation 4.2
S
y
(, ) =
n
i=1
S
y
i
i
H
g
(, ) =
n
i=1
H
g
i
i
(, ) =
n
i=1
i
(4.2)
where S
y
i
, H
g
i
, and
i
are the yield stress, gap, and error respectively at node i. Thus, when the
nine simulations are run, the error can be expressed over the space dened by the values of S
y
and H
g
.
4.3.2 Error Function
The error for the yield-gap calibration requires specic denition. The errors being referred to are
percent errors between the simulation predicted values and test data values of exit thickness H
X
and rolling load P. Thus, the percent error
p
and
h
are dened in Equation 4.3.
p
=
_
Predicted P Actual P
Actual P
_
100
h
=
_
Predicted H
X
Actual H
X
Actual H
X
_
100 (4.3)
This gives two errors:
p
for rolling-load error and
h
for exit-thickness error. The total error
t
used in the response surface is dened by Equation 4.4;
t
=
2
p
+
2
h
(4.4)
thus, the minimum total error represented by the surface will be zero.
4.3.3 Simulation Parameters
The nine simulations represent a full factorial experiment with three values of yield stress and
three values of gap. The test data gave a yield stress of 0.02119 tonne/mm
2
and gap of 3.98 mm.
New values for yield stress and gap were chosen within approximately ten percent of the original
values. The calibration values for yield stress were 0.019, 0.021, and 0.023 tonne/mm
2
and the
values for gap were 3.1, 3.5, and 3.9 mm. The experimental design is summarized in Table 4.6.
4.3.4 Simulation Generation & Running
The simulations were generated using the 3-D rolling model. The data used to construct that
model are listed in Section 4.2.3 except for the yield stress and gap values which are given in Table
4.6. A total of nine simulations were generated and run.
77
Table 4.6: Yield-Gap Calibration design.
Node S
y
H
g
Node S
y
H
g
# tonne/mm
2
mm # tonne/mm
2
mm
1 0.019 3.1 6 0.023 3.5
2 0.021 3.1 7 0.019 3.9
3 0.023 3.1 8 0.021 3.9
4 0.019 3.5 9 0.023 3.9
5 0.021 3.5
4.3.5 Data Extraction & Results
The data was extracted from the output databases. The following graphs are presented for ob-
serving simulation quality, so they do not reference specically which node each simulation rep-
resents.
Figure 4.2 shows the rolling load versus time for the nine simulations. The rolling-load curves
all have a similar shape and atten out to steady-state at approximately the same time. Low error
variance is also shown in the steady-state regions. The exit thickness along the length of the strip
is shown in Figure 4.3. While the exit-thickness curves are not as clean as the rolling-load curves,
they do all show the system reaching steady state. The exit thickness versus strip width with
curve ts is shown in Figure 4.4. These curves show that the parabola t does well represent the
strip crown over a variety of conditions. The spread around each curve also indicates the quality
of the steady-state data.
Figure 4.2: Rolling load versus time for yield-gap calibration simulations
78
Figure 4.3: Exit thickness versus strip length for yield-gap calibration simulations.
The numerical data extracted from the graphs is summarized in Table 4.7. The symbols S
y
,
H
g
, P, and H
X
represent the yield stress, gap, rolling load, and exit thickness respectively. The
exit-thickness error, rolling-load error and total error are given by
h
,
p
, and
t
.
Table 4.7: Yield-gap calibration simulation data summary.
Node S
y
H
g
P H
X

g

p

t
# tonne/mm
2
mm tonne mm % % %
1 0.019 3.1 1164.82 5.2032 5.20 -5.34 55.63
2 0.021 3.1 1227.9 5.306 10.90 -3.47 130.90
3 0.023 3.1 1281.87 5.4013 15.78 -1.74 251.90
4 0.019 3.5 1041.22 5.3944 -5.96 -1.86 38.99
5 0.021 3.5 1092.42 5.4821 -1.33 -0.27 1.85
6 0.023 3.5 1141.42 5.5698 3.09 1.33 11.31
7 0.019 3.9 911.43 5.5892 -17.68 1.68 315.46
8 0.021 3.9 961.34 5.6604 -13.17 2.97 182.40
9 0.023 3.9 1008.29 5.7384 -8.93 4.39 99.11
4.3.6 Calibration Results
The values of yield stress S
y
, gap H
g
, and the error (
p
,
h
, and
t
) were used to construct re-
sponse surfaces. Figure 4.5 shows the response surface represented as a contour of rolling-load
79
Chapter 4. Calibration, Validation, & Sensitivity 4.4. Work Roll Crown Calibration
Figure 4.4: Exit thickness versus strip width for yield-gap calibration simulations.
error squared
2
p
such that the minimum error was zero. A valley of zero error is shown going
diagonally upward from left to right in the contour.
The strip exit-thickness error squared
2
h
is shown in Figure 4.6. Once again the contour rep-
resents the error squared such that all error is positive and the minimum error is zero. Another
valley of zero error is found going diagonally downward from left to right.
The total error contour
t
represents the sum of the squares of rolling-load and strip exit-
thickness error, is shown in Figure 4.7. When the contours from Figures 4.5 and 4.6 are combined,
the valleys merge to create a localized minimum of error. The red marker at the center repre-
sents the minimum error in the contour. The location of this point was found using a bounded
minimizer in the python scripting language given the error represented by the response surface.
Thus, the minimum error occurs when the yield stress is 0.02145 tonne/mm
2
and the gap is
3.489 mm. The calibrated yield stress and gap are 1.23%and 12.40%percent different fromthe test
data respectively.
4.4 Work Roll Crown Calibration
The purpose of the work roll crown calibration was to adjust for the thermal expansion of the roll
affecting its crown. The actual thermal expansion was not modeled, but its effects were simulated
by changing the work roll crown C
w
. Since only one variable is being varied, fewer runs are
necessary to create a repsonse surface. In fact, the response surface is one dimensional in the
work roll crown calibration.
80
Figure 4.5: Rolling-load error contour,
2
p
.
Figure 4.6: Exit-thickness error contour,
2
h
.
81
Figure 4.7: Total error contour,
t
=
2
p
+
2
h
.
4.4.1 Response Surface
The response surface denition for this calibration was different than that of the yield-gap cali-
bration. Since more data points could easily be taken and only one parameter was being varied,
a curve t of the data was used to generate the response surface function. Five simulations were
run with varying work roll crowns and then the exit crown error
c
was calculated in reference to
the measured strip crown as given by Equation 4.5
c
=
_
Predicted C
X
Actual C
X
Actual C
X
_
100 (4.5)
where C
X
is the exit crown.
4.4.2 Data Extraction & Results
The test data gave a work roll crown of -0.16 mm. Simulations were run with the work roll crowns
shown in Table 4.8:. The curve ts of the strip data to determine the crowns are shown in Figure
4.8. The graph shows that changes in work roll crown have little effect on the exit thickness at the
centerline of the strip. There is more variation in this data near the edge of the strip. The work roll
crowns and corresponding exit crowns predicted by the simulations are summarized in Table 4.8.
4.4.3 Calibration Results
The strip crown error squared
2
c
was plotted against the work roll crown C
w
as shown in Figure
4.9. The data was t with a second order polynomial described by Equation 4.6.
82
Figure 4.8: Exit thickness versus strip width for work roll crown calibration.
Table 4.8: Roll crown calibration simulation data summary.
Sim C
w
C
X

c

2
c
# mm mm % %
1 -0.16 0.1297 310.4 96374.9
2 -0.1 0.0958 203.1 41275.8
3 -0.04 0.061 93.0 8656.1
4 0.02 0.0287 -9.2 84.2
5 0.08 -0.007 -122.2 14921.1
2
c
= a
1
C
2
w
+a
2
C
w
+a
3
(4.6)
The curve t yielded the constants in Equation 4.7.
a
1
= 3.2524e + 06 a
2
= 7.9975e + 04 a
3
= 4.4250e + 02 (4.7)
The roots of the t equation were found to be 0.0162 and 0.0084. These roots were averaged to
give a calibrated work roll crown of 0.0123 mm. The calibrated crown is signicantly different
from the original crown. However, the work roll crown has been shown to increase by as much as
0.5 mm in the rst minute of rolling a similarly sized strip [9].
83
Chapter 4. Calibration, Validation, & Sensitivity 4.5. Validation
Minimum
Figure 4.9: Strip crown error versus work roll crown.
4.5 Validation
Two validations were performed, one after each calibration. This was done to determine the
effectiveness of each calibration.
4.5.1 Yield-Gap Calibration Validation
The yield-gap calibration validation was performed using 0.02145 tonne/mm
2
for the yield stress
and 3.489 mm for the gap. The 3-D rolling model was run to verify that the calibration was able
to reduce the error in rolling load and exit thickness. The results of this simulation produced a
rolling load of 1108.21 tonnes and an exit thickness of 5.4984 mm. These predicted rolling loads
and exit thicknesses differ from the measured test data of 1107.20 tonnes and 5.4969 mm by less
than 0.01%. Thus, the calibration was successful in reducing the error in the parameters for which
the error was measured. The strip crown has not improved as the validation predicts a crown of
0.1297 mm, a difference of 310% from the test data. This data is summarized in Table 4.9.
Table 4.9: Yield-gap calibration validation data summary.
Exit Thickness H
X
mm 5.4984 5.4969 0.027
Strip Crown C
X
mm 0.12970 0.0316 310
84
Chapter 4. Calibration, Validation, & Sensitivity 4.6. Sensitivity Analysis
4.5.2 Final Calibration Validation
The calibrated work roll crown, 0.0123 mm, combined with the yield-gap calibration results, was
used to generate and run a simulation. This validation returned a rolling load of 1083.26 tonnes,
an exit thickness of 5.4950 mm, and an exit crown of 0.0330 mm. The validation results, test data,
and percent error between them is summarized in Table 4.10. All of the predicted values are less
than ve percent different from the test data.
Table 4.10: Final validation data summary.
Exit Thickness H
X
mm 5.495 5.4969 0.035
Strip Crown C
X
mm 0.0330 0.0316 4.430
4.6 Sensitivity Analysis
Nearly every aspect of the rolling mill stand affects the strip exit crown. However, most of these
parameters are dictated by product requirements other than atness. Therefore, gap, strip entry
crown, work roll crown, and bending force were chosen as the parameters to study. The gap is set
at each stand in a nishing mill to produce a product of a given thickness, but this could be used
to control strip crown as well. The strip entry crown is the exit crown from the previous stand. By
controlling the successive exit crowns, it is possible to control the strip entry crown. The work roll
crown can be controlled not only by grinding a certain prole, but also by the manner in which
the roll is cooled. By controlling the initial shape of the work roll crown and its temperature, it
becomes possible to control the work roll crown itself. Finally, the bending force is specied by
the mill controller for the purpose of changing strip crown. Its effectiveness at controlling the strip
crown can be found directly through sensitivity analysis.
4.6.1 Parameter Generation
To study the effects on strip crown, each of the selected parameters were varied above and below
the original value by 20% in 5% increments. The values of the parameters are summarized in
Table 4.11. Simulations were generated for each perturbation for a total of 36 simulations.
4.6.2 Results
The 36 simulations were run and data was extracted. Plots were generated of roll crown versus
each parameter and a linear t was made of each data set. Figures 4.10, 4.11, 4.12, and 4.13 show
the data and curve ts for gap, entry crown, roll crown, and bending force respectively. Exit crown
versus gap, Figure 4.10, shows a strong correlation and a good t with a R
2
value of 0.991. Exit
crown versus entry crown, shown in Figure 4.11, has poor correlation due to the amount of scatter
in the data. The curve t has a R
2
value of 0.570 indicating a poor t. The data for exit crown
versus work roll crown had a poor t as well. There are more robust estimation techniques that
can be applied to the entry crown and work roll crown data in the future. Lastly, the exit crown
versus bending force shows a strong correlation with a R
2
value of 0.970. Further analysis should
be conducted to determine the source of variation in the work roll crown and entry crown results.
85
Table 4.11: Sensitivity analysis parameter perturbation summary.
Sim Multiplier H
g
[mm] C
E
[mm] C
w
[mm] J [tonne]
1 0.80 2.791 0.0389 0.009836 77.5
2 0.85 2.966 0.0413 0.010451 82.35
3 0.90 3.140 0.0437 0.011100 87.19
4 0.95 3.315 0.0462 0.011700 92.04
5 1.00 3.489 0.0486 0.012300 96.88
6 1.05 3.664 0.0510 0.012900 101.72
7 1.10 3.838 0.0535 0.013500 106.57
8 1.15 4.013 0.0559 0.014139 111.41
9 1.20 4.187 0.0583 0.014753 116.26
Figure 4.10: Strip exit crown versus gap.
86
Figure 4.11: Strip exit crown versus strip entry crown.
Figure 4.12: Strip exit crown versus work roll crown.
87
Figure 4.13: Strip exit crown versus bending force.
The slope of each curve t is summarized in Table 4.12. The sensitivities for entry crown and
work roll crown may not be very accurate due to the poor quality of the curve t for those data
sets. Changes in work roll crown and work roll crown appear to have the greatest effect on the
exit crown. The gap has little effect on the exit crown. The bending force, the main method by
which the crown is controlled, appears to have little effect on the exit crown.
Table 4.12: Sensitivity analysis results.
Parameter Symbol Sensitivity
Gap H
g
-0.0185 mm/mm
Strip Entry Crown C
E
0.2198 mm/mm
Work Roll Crown C
w
-0.9735 mm/mm
Bending Force J -0.0003 mm/tonne
Upon careful examination, it is possible to see that the results from varying the strip entry
crown and the work roll crown may be noise. The range of strip exit crowns produced by these
analyses are signicantly smaller than those produced by varying the gap or the bending force.
A wider range of strip entry crowns and work roll crowns may yield data with a more denite
trend.
88
Chapter 4. Calibration, Validation, & Sensitivity 4.7. Summary
4.7 Summary
Two calibrations were performed, one to minimize the error in exit thickness and rolling load,
and the other to minimize exit crown error. Validations were performed after each calibration.
The yield-gap calibration reduced the rolling-load and exit-thickness errors to less than 1%. The
work roll crown calibration decreased the exit crown error from 275% to less than 5%. Thus, the
calibrated model predicts rolling load, exit thickness, and strip exit crown within 5% of test data.
Sensitivity analysis was performed to determine how gap, entry crown, work roll crown, and
bending force affect the strip exit crown.
89
5 Summary, Conclusions, and Recommendations
This chapter presents a summary of the work performed for this thesis, conclusions that may be
drawn from it, and recommendations relating to further work.
5.1 Summary
Many models and methods were developed in the course of creating a rened dynamic roll-stack
model for the hot-rolling process. The process began with model development based on the pro-
vided test data from a stand in an actual hot-rolling mill. Four types of models were constructed
including the nal 3-D rolling model which was the goal of this project. The next step was to
calibrate the 3-D rolling model to account for possible errors in the test data. Once the nal model
was calibrated, a validation simulation was performed to determine the ability of the model to
predict the results in the test data. Lastly, a sensitivity analysis was performed to determine the
effects of four stand parameters on strip crown.
5.1.1 Model Development
Model development began by dening the strip material properties. An elastic-plastic model was
used with strain hardening represented as an elliptical function of strain and the yield stress. This
model was used to allow any metal material to be modeled, not just steel. This choice allows the
nite element models which use this material model to be used for any hot-rolling application
whether the material be aluminum, stainless steel, or any other metal.
Finite element modeling is dependent on the representation of the physical system. Therefore,
the stand and roll-stack arrangement was carefully examined to nd methods of simplication.
Due to symmetry, the stand was decomposed such that a one-quarter-symmetry model could be
used to create a dynamic rolling model and a one-eighth symmetry model could represent a static
stiffness model. It was also found that the stand frame could be replaced by a spring of equivalent
stiffness which greatly simplied model construction because modeling the frame geometry was
not required.
The next step was to determine the stiffness of the roll-stack by creating the Roll-Stack Stiffness
Model. This model was implemented in the nite element software Abaqus in both two and three
dimensions. The Roll-Stack Stiffness Model compressed the roll-stack in a manner similar to the
stand calibration performed at the mill. A load-displacement curve was generated by the model
from which the stiffness of the roll-stack was extracted. From the roll-stack stiffness the stiffness
of the frame was predicted.
The Frame Stiffness Model was developed to ascertain the accuracy of the roll-stack models
frame stiffness prediction. This model represented the roll-stack with the backup roll supported
by a spring representing the frame. When the roll-stack was compressed, so was the spring, which
simulates the stand calibration in the mill. Aload-displacement curve was generated and the total
stiffness of the roll-stack and frame was found. The total stiffness of the frame and roll-stack was
less than 0.5% different from the total stand stiffness determined in the test data.
Whereas the previous two models had dealt with the static behaviour of the roll stack, the
Lumped-Mass Model was created to determine lumped parameters to be used to represent damp-
90
Chapter 5. Summary, Conclusions, and Recommendations 5.1. Summary
ing elements in the 2-D and 3-D dynamics nite element models. The lumped-mass model was
an explicit nite element model developed in the Python scripting language to nd the frequency
of vibration of the rolls and damping coefcients to bring the roll-stack to steady state. The model
relied on the estimates of the frame, roll-stack, and strip stiffnesses along with rolling load pre-
dictions based on the Ford and Alexander rolling load model. This model was used to determine
multipliers to set run time and strip displacement time. The multipliers were then used in the 2-D
and 3-D Rolling Models to set how long the nite element simulations would run.
The Dynamic 2-D Rolling Model was developed. This was a nite element model created in
Abaqus to simulate the rolling of a strip. A plane strain stress-strain relationship was assumed
and the previously mentioned elastic-plastic material model was used. This model was used for
exploration of rolling modeling and meshing techniques for nding a reliable solution. Once this
model was veried to work, the 3-D rolling model was constructed.
The dynamic 3-D rolling model was the nal model developed. It was able to calculate rolling
load, exit thickness, and strip crown. This model was capable of representing all of the important
mechanics of the rolling process except for heat transfer. A mesh convergence study was per-
formed to determine how the mesh affected the solution quality. Finally, a simulation was run to
determine how well the 3-D rolling model replicated the measurements in the test data. The exit
thickness was 3.79% different from the test data and the rolling load was 15.39% different. The
strip exit crown had a percent error of 275% from the test data. A calibration procedure was later
applied to reduce test data uncertainties that could have been the source of these errors.
5.1.2 Calibration
The stand parameter calibrations were used to reduce test data error and to minimize model
prediction error. The test data and its collection methods were carefully examined for possible
sources of error. It was found that the most likely sources of error were in the strip yield stress,
gap, and work roll crown measurements. Two calibrations were performed, one to minimize exit
thickness and rolling load error, the other to minimize strip exit crown error.
The yield-gap calibration varied the strip yield stress and the gap to nd a combination which
would minimize the rolling load and exit thickness error. Nine simulations were run representing
a full-factorial two-parameter three-level expansion of three yield stresses and three gaps. From
the results, a response surface of total rolling-load and exit-thickness error was constructed using
quadratic isoparametric interpolation functions. The error over the sample space was minimized
to give the values of yield stress and gap that would minimize the error. This calibration produced
a yield stress of 0.02145 tonne/mm
2
and a gap of 3.489 mm. The calibrated yield stress differed
from the test data yield stress of 0.02119 tonne/mm
2
by 1.23% and the predicted gap differed
from the test data value of 3.983 mm by 12.40%. These calibrated values were used in the work
roll crown calibration and in the model validation.
The work roll crown calibration attempted to account for the effects of the thermal expansion
on the work roll crown, which affects the strip crown. Five simulations were run with varying
work roll crowns. The error in the strip crown versus the test data was computed and graphed.
The work roll crown that minimized the error was chosen. The calibration gave a work roll crown
of 0.0123 mm which differed from the test data value of -0.16 mm by 107%. While the difference
seems signicant with respect to the percent difference, the actual change in crown was only
0.1477 mm and accounts for all thermal expansion of the work roll while rolling the strip.
91
Chapter 5. Summary, Conclusions, and Recommendations 5.1. Summary
5.1.3 Validation
Two validations were performed, one after each calibration, to determine their effectiveness and
the nal quality of the 3-D rolling model. The validation performed after the yield-gap calibration
showed that the exit thickness and rolling load error had decreased below 0.01%. Therefore, the
yield-gap calibration was successful at reducing error. However, those values of yield stress and
gap caused the strip crown error to reach 310%. The rolling load P, strip exit thickness H
X
, and
strip crown C
X
predictions and errors for the validation performed after the yield-gap calibration
are presented in Table 5.1.
Table 5.1: Yield-gap calibration validation data summary.
Variable Unit Predicted Actual % Error
P tonne 1108.21 1107.2 0.091
H
X
mm 5.4984 5.4969 0.027
C
X
mm 0.12970 0.0316 310
The nal validation was performed with the data taken from both the yield-gap and work roll
crown calibrations. The results of the nal validation showed that the work roll crown calibration
had caused the strip exit crown error to have dropped from 312% to less than 5%. However, that
calibration also caused an increase in the rolling load and exit thickness error, though they both
errors remained at less than 3%. Therefore, the nal calibrated dynamic 3-D rolling model was
able to exceed the goal of matching test data measurements within 10% error. The rolling load
P, strip exit thickness H
X
, and strip crown C
X
predictions and errors for the nal validation are
presented in Table 5.2.
Table 5.2: Final validation data summary.
Variable Unit Predicted Actual % Error
P tonne 1083.26 1107.2 2.162
H
X
mm 5.495 5.4969 0.035
C
X
mm 0.0330 0.0316 4.430
The nal validation had greater rolling-load and exit-height errors than the yield-gap calibra-
tion validation. This effect was most likely due to changes in work roll crown affecting the rolling
load and exit height. The work roll crown calibration does not appear to have been as successful
at reducing error as the yield-gap calibration, which reduced errors to less than 0.1%. This may
be due to the highly sensitive nature of the strip crown and the small values involved, less than
0.2 mm.
5.1.4 Sensitivity
The sensitivity analysis used the dynamic 3-D rolling model to determine which of several stand
parameters affect the strip crown the most. The most controllable aspects of the stand were chosen
as parameters for the sensitivity analysis: gap, strip entry crown, work roll crown, and bending
force. Nine simulations were run for each parameter with varying values above and below the
calibrated model. The exit crowns results were plotted versus the parameters and then curve t.
The slopes of the linear ts were determined and compared. While the entry crown and work
92
Chapter 5. Summary, Conclusions, and Recommendations 5.2. Conclusions
roll crown results showed little correlation and bad t quality, the gap and bending force data
showed good t quality. The poor ts may have been due to the narrow range of variations used
to perturb the system. By using a small range of work roll crowns and entry crowns, the exit
crown did not vary sufciently to well represent the solution space. Additional simulations need
to be run to determine the sensitivities conclusively due to the poor quality of the curve ts and
the narrow range of the data. Table 5.3 summarizes the sensitivity of strip exit crown to gap, strip
entry crown, work roll crown, and bending force.
Table 5.3: Sensitivity analysis results.
Parameter Symbol Sensitivity
Gap H
g
-0.0185 mm/mm
Strip Entry Crown C
E
0.2198 mm/mm
Work Roll Crown C
w
-0.9735 mm/mm
Bending Force J -0.0003 mm/tonne
5.2 Conclusions
There are many conclusions that can be drawn from the work presented in this thesis. The more
important topics include the nal models predictive ability, calibration, and performance versus
effort.
5.2.1 Predictive Ability
The dynamic 3-D rolling model was very successful at predicting the parameters it set out to
predict. It was capable of predicting the rolling load, exit thickness, and strip crown within 5% of
the test data measurements, less than half of the error specied in the objectives. The rolling load
and exit thickness predictions were within 0.5% of the test predictions. Further renements of the
model could bring the strip-crown error down to that of the exit-height and rolling-load error.
The sensitivity analysis results are also encouraging. Varying gap and bending force showed
pronounced effects on the strip crown. The quality of the curve ts, R
2
values of 0.991 and 0.970
for the gap and bending force sensitivities, indicate that using this analysis to predict strip crown
is possible. Further model testing and renements could reduce the noise in the strip entry crown
and work roll crown sensitivities. In addition, these results clearly show how little an effect bend-
ing force has on strip crown. The information from the sensitivity analysis will be useful in deter-
mining which parameters would be most effective at controlling strip crown.
5.2.2 Calibration
The yield-gap calibration involved modifying the yield stress and gap by less than 15%. The work
roll crown calibration changed the roll crown by less than 0.2 mm to bring the strip exit crown
error from 310% to within 5% of the measured value. The relatively small adjustments of the
calibration parameters show that not only is rolling load, exit thickness, and strip crown very
sensitive, but also that the nal parameters were not signicantly different from the test data.
93
Chapter 5. Summary, Conclusions, and Recommendations 5.3. Recommendations
5.2.3 Performance versus Effort
The ability to predict the strip prole within 5% of test data is a step forward in modeling the
rolling process; however, it does have a cost. Each of the models developed in this thesis is re-
quired to simulate a stand. The roll-stack stiffness model is needed to predict the stand frame stiff-
ness. The frame stiffness model must be used to validate the frame stiffness. The lumped-mass
model is needed to determine run time parameters and damping coefcients. The 2-D rolling
model is used to verify those coefcients and conrm that the model reaches steady state. Fi-
nally, the 3-D model requires 14 runs to complete the calibration. There is much effort required to
generate and calibration the 3-D rolling model; however, the effort is worthwhile.
Without all of these models and steps, it would not be possible to accurately simulate the
rolling process. The results indicate predictions within 5% of test data. This predictive ability is
signicant considering that the rolling loads are over 1100 tonnes and the strip crowns are less
than 0.05 mm.
Though the 3-D rolling model requires approximately an hour per simulation, advances in
computing technology will continually reduce run times. Since optimizing run times was not a
goal of this project, the focus was more on quality of solution than speed. Further studies into the
models behaviour may yield simplications or adjustments that signicantly reduce run times,
making it easier to explore the rolling process.
5.3 Recommendations
The following recommendations are based on observations and conclusions made during the
course of the project. Areas where further development should occur include model develop-
ment, sensitivity analysis, process exploration, and crown equation.
5.3.1 Model Development
While the dynamic 3-D rolling model represents an advancement in the modeling of rolling, there
remains roomfor improvement. Modeling of the heat transfer between the strip and rolls could be
used to increase the accuracy of the model prediction. This could also be used to develop schedul-
ing and cooling methods to predict roll thermal expansion and introduce another method of con-
trolling crown. Another area to explore in model development is reducing simulation solution
times. At approximately an hour per simulation with the dynamic 3-D rolling model, performing
a large number of runs quickly becomes prohibitive. By further studying mesh and representa-
tion tradeoffs, it may be possible to signicantly reduce solution times. For example, using the
3-D Rolling Model as a baseline, the effectiveness of a static 3-D nite element model could be
explored. If it were possible to achieve similar results with the simpler static model, simulation
times could be reduced signicantly.
5.3.2 Process Exploration
With a model available to represent so many aspects of the rolling process, it becomes possible
to perform controlled experiments in the rolling process. Each stand parameter can be varied
independently and the resulting effects on the strip can be measured. This could lead to a level
of control not previously possible. For instance, the exact effects of changes in rolling friction on
rolling load and exit crown could be determined. The effects of strip prole representation could
be explored. The 3-D rolling model opens up many possibilities for the advancement of atness
control.
94
Chapter 5. Summary, Conclusions, and Recommendations 5.3. Recommendations
5.3.3 Sensitivity Analysis
The sensitivity analysis presented in this work was only a preliminary study. The principles used
here could be applied to any stand or strip parameter. The strip width and roll sizes are other
parameters that heavily inuence strip crown and can be controlled during mill setup. The effect
of frame stiffness on strip crown and atness may be of interest to those designing the frames.
In addition, developing sensitivities to parameters at each stand in a mill would allow the mill
controllers to determine how much to change the crown and thickness at each stand to achieve a
at nal product. With this model, it would be possible to design an entire framework to control
the mill based on changing the most inuential parameters to minimize the strip atness error.
5.3.4 Crown Equation
With further developments in sensitivity analysis, it would become possible to create a crown
equation based on Taylor series expansion. A set of controllable parameters which affect the strip
crown could be chosen and perturbations around a base conguration could be performed to
nd sensitivities. This could be done for each conguration of each mill stand so that the online
controllers could predict strip crown without the need to run a model.
95
Bibliography
[1] Abaqus. Abaqus Online Documentation: Version 6.8. Dassault Systemes, 2008.
[2] J. M. Alexander. On the theory of rolling. Proceedings of the Royal Society of London. A. Mathe-
matical and Physical Sciences, 326(1567):535563, 1972.
[3] Robert D. Cook, David S. Malkus, Michael E. Plesha, and Robert J. Witt. Concepts and Appli-
cations of Finite Element Analysis. John Wiley and Sons, Inc., 4th edition, 2002.
[4] George Dieter. Mechanical Metalurgy. McGraw-Hill, 3rd edition, 1986.
[5] Eduardo N. Dvorkin, Marcela B. Goldschmit, Miguel A. Cavaliere, Pablo M. Amenta, Os-
valdo Marini, and Walter Stroppiana. 2d nite element parametric studies of the at-rolling
process. Journal of Materials Processing Technology, 68(1):99 107, 1997.
[6] I. J. Freshwater. Simplied theories of at rollingi. the calculation of roll pressure, roll force
and roll torque. International Journal of Mechanical Sciences, 38(6):633 648, 1996.
[7] I. J. Freshwater. Simplied theories of at rollingii. comparison of calculated and experi-
mental results. International Journal of Mechanical Sciences, 38(6):649 660, 1996.
[8] Vladimir B. Ginzburg. Steel Rolling Technology: Theoy and Practice. Marcel Dekker, Inc., New
York, 1989.
[9] Vladimir B. Ginzburg. High-Quality Steel Rolling: Theory and Practice. Marcel Dekker, Inc.,
1993.
[10] Remn-Min Guo. Prediction of strip prole in rolling process using inuence coefcients and
boussinesqs equations. Journal of Manufacturing Science and Engineering, 119(2):220226, 1997.
[11] A. Hacquin, P. Montmitonnet, and J. P. Guillerault. A three-dimensional semi-analytical
model of rolling stand deformation with nite element validation. European Journal of Me-
chanics - A/Solids, 17(1):79 106, 1998.
[12] Daniel J. Inman. Engineering Vibration. Prentice Hall, 3rd edition, 2007.
[13] J. G. Lenard, M. Pietrzyk, and L. Cser. Mathematical and Physical Simulation of the Properties of
Hot Rolled Products. Elsevier, Oxford, 1999.
[14] John G. Lenard. Primer on Flat Rolling. Elsevier, 2007.
[15] Zone-Ching Lin and Ven-Huei Lin. A study of radial thermal expansion of a work roll for
three-dimensional strip deformation during rolling. Journal of Materials Processing Technology,
70(1-3):62 76, 1997.
[16] Daryl L. Logan. A First Course in the Finite Element Method. Thomsom, 4th edition, 2007.
[17] Arif S. Malik and Ramana V. Grandhi. A computational method to predict strip prole in
rolling mills. Journal of Materials Processing Technology, 206(1-3):263 274, 2008.
96
Bibliography
[18] K. Mori, K. Osakada, and T. Oda. Simulation of plane-strain rolling by the rigid-plastic nite
element method. International Journal of Mechanical Sciences, 24(9):519 527, 1982.
[19] Uwe Niggemeier. Stahlseite: Industriefotograe, 2009.
[20] E. Orowan. The calculation of roll pressure in hot and cold at rolling. Proceedings of the
Institution of Mechanical Engineers, 150:140167, 1943.
[21] Maciej Pietrzyk and John G. Lenard. Thermal-Mechanical Modeling of the Flat Rolling Process.
Springer-Verlag, Germany, 1991.
[22] J. N. Reddy. An Introduction to the Finite Element Method. McGraw-Hill, New York, NY, 3rd
edition, 2006.
[23] William L. Roberts. Hot rolling of steel. Manufacturing engineering and materials processing.
Marcel Dekker, Inc., 270 Madison Avenue, New York, New York 10016, 1983.
[24] William L. Roberts. Flat Processing of Steel. Manufacturing engineering and materials pro-
cessing. Marcel Dekker, Inc., 270 Madison Avenu, New York, New York 10016, 1988.
[25] Harold Rothbart. Cam Design Handbook. McGraw-Hill, 2004.
[26] K. Saxl. Transverse gauge variation in strip and sheet rolling. Proceedings of the Institution of
Mechanical Engineers, 172:727742, May 1958.
[27] K. N. Shohet and N. A. Townsend. Roll bending methods of crown control in four-high plate
mills. Journal of The Iron and Steel Institute, pages 10881098, November 1968.
[28] R. B. Sims. The calculation of roll force and torque in hot rolling mills. Proceedings of the
Institution of Mechanical Engineers, 168:191200, 1954.
[29] Simulia. Abaqus 6.8-2, 2008.
[30] Peter Wriggers. Nonlinear Finite Element Methods. Springer Berlin Heidelberg, 2008.
[31] Zygmont Wusatowski. Fundamentals of Rolling. Pergamon Press, London, 1969.
[32] Ki Ho Yun, Tae Jin Shin, and Sang Moo Hwang. A nite element-based on-line model for the
prediction of deformed roll prole in at rolling. ISIJ International, 47(9):13001308, 2007.
97
Appendix: Test Data
98
Appendix: Test Data
Table 1: Work roll data.
Property Symbol Values Units
Barrel Length B
lw
2080 mm
Base Diameter B
dw
655.715 mm
Neck Length N
lw
195 mm
Neck Diameter N
dw
510 mm
w
-0.16 mm
Shell Thickness t
sw
50 mm
cw
21.0920874 tonne/mm
2
Core Poissons ratio
cw
0.29
sw
21.0920874 tonne/mm
2
sw
0.29
Density
w
7.96E-13 tonne-s
2
/mm
4
Table 2: Backup roll data.
Barrel Length B
lb
1820 mm
Base Diameter B
db
1589.505 mm
Neck Length N
lb
360 mm
Neck Diameter N
db
945 mm
b
0 mm
Shell Thickness t
sb
60 mm
cb
21.0920874 tonne/mm
2
Core Poissons ratio
cb
0.29
sb
21.0920874 tonne/mm
2
sb
0.29
Density
b
7.96E-13 tonne-s
2
/mm
4
99
Appendix: Test Data
Table 3: Strip data.
Width w 1524.84 mm
Feather w
f
40 mm
Entry Thickness H
E
6.4337 mm
Exit Thickness H
X
5.4969 mm
Entry Crown C
E
0.0486 mm
Exit Crown C
X
0.0316 mm
Elastic Modulus E
s
18 tonne/mm
2
Yield Stress S
y
0.02118574 tonne/mm
2
Sy Multiplier 2
Poissons Ratio
s
0.3
Density
s
7.96E-13 tonne-s
2
/mm
4
Temperature T 853.35 Celsius
Velocity v 6752.64 mm/s
Table 4: Stand data.
Rolling Load P 1107.2 tonne
Bending Force J 96.88 tonne
b
150 tonne
Gap H
g
3.9831 mm
Strip-Roll Friction
s
0.2
Roll-Roll Friction
r
0.2
Total stiffness k
t
610 tonne/mm
100
Appendix: Rolling Load Model
# Ro l l i ng Load Model
# Based on Ford and Al e xande r
#
# Func t i ons :
# ge t Load ( k , he , hx , w, D, E, v )
# get Kx ( k , he , hx , D, E, v )
#
# Common I nput s :
# k f l o w s t r e s s , Sy / s q r t ( 3 )
# he s t r i p e nt r y h e i g h t
# hx s t r i p e x i t h e i g h t
# w s t r i p wi dt h
# D work r o l l di a me t e r
# E work r o l l e l a s t i c modul us
# v work r o l l Po i s s o n s r a t i o
from numpy import pi , s qr t
def getLoad ( k , he , hx , w, D, E, v ) :
Output : P t o t a l r o l l i n g l o a d
dh = he hx # d r a f t
hAvg = 0 . 5 ( he + hx )
Rp = D/2. # i n i t i a l i z e r a di us
C = 16( 1 v2) /( pi E) # r o l l de f o r ma t i o n c o ns t a nt
pr i nt C
for i in range ( 1 0 ) : # c o nve r ge l o a d and de f o r ma t i o n
arc = s qr t ( Rpdh)
F = k arc ( pi 0. 5 + 0. 5 arc/hAvg) # uni t r o l l i n g l o a d
Rp = D/2. ( 1 + CF/dh) # d e f l e c t e d r o l l r a di us
P = Fw
ret urn P # r e t ur n t o t a l r o l l i n g l o a d
def getKx ( k , he , hx , D, E, v ) :
Output : Kx s t r i p s t i f f n e s s
hxa = hx 1. 025 # e x i t h e i g h t a
hxb = hx 0. 975 # e x i t h e i g h t b
Pa = getLoad ( k , he , hxa , 1 , D, E, v) # r o l l i n g l o a d a
Pb = getLoad ( k , he , hxb , 1 , D, E, v) # r o l l i n g l o a d b
Kx = abs ( ( Pb Pa ) /( ( hxb hxa ) ) )
ret urn Kx
101

Developing a Refined Dynamic Roll-Stack Model for Hot Rolling

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Developing a Refined Dynamic Roll-Stack Model for Hot Rolling

Hochgeladen von

Copyright:

Verfügbare Formate

Strip Crown Prediction: Developing a Rened Dynamic

Roll-Stack Model for the Hot Rolling Process

Change in strip crown within a stand [mm]

Change in strip centerline thickness within a stand [mm]

Work roll radius deformed [mm]

, called draft, is the difference between the

within a single stand.

is the change in thickness of the strip [9].

is the deformed roll radius, H

is the change in strip thickness, E is the elastic modulus

Das könnte Ihnen auch gefallen